个人知识图谱对个人数据社区和知识图谱社区都做出了重要贡献。许多人使用个人和笔记应用程序，却没有意识到它们在更广泛的数据生态系统中存在联系的潜力。到目前为止，知识图谱社区主要关注通过关联数据计划使大型公共数据集可连接，或通过企业知识图谱链接各个公司的数据孤岛。这本书架起了一座可以双向穿越的桥梁。
Personal Knowledge Graphs is an important contribution, both to the Personal Data community as well as to the Knowledge Graph community. Many people use personal and note-taking applications without being aware of the potential for their connection in a much broader ecosystem of data. Up until now, the Knowledge Graph community has been primarily focused either on making large public datasets connectable through Linked Data initiatives, or linking the data silos of individual firms through Enterprise Knowledge Graphs. This book lays down a bridge that can be traversed both ways.
Dave McComb，Semantic Arts, Inc. 总裁兼联合创始人
Dave McComb, President & Co-founder, Semantic Arts, Inc.
知识图谱是贯穿人工智能的数据和知识管理的有效范例。个人知识图谱的兴起可以赋予个人数据主权和控制权。本书是第一本探讨这一热门话题的书。
Knowledge graphs are an effective paradigm for data and knowledge management permeating AI. The rise of Personal Knowledge Graphs can empower individuals towards data sovereignty and control. This book is the first to explore this timely topic.
比纳·阿玛纳特 (Beena Ammanath)，《值得信赖的人工智能》作者
Beena Ammanath, author of Trustworthy AI
知识图谱目前已被业界和政府广泛接受，是一种组合、存储和查询大量异构数据的有效方法。本书首次为知识图谱的新应用打开了大门：希望控制自己数据的个人公民，其应用范围从个人存档一直到个人数字助理。本书汇集了一系列易于理解的贡献，为个人知识管理的新愿景打开了大门。
Knowledge Graphs are now widely accepted in industry and government as an effective way to combine, store and query large volumes of heterogeneous data. This book is the first to open the door to a new application of knowledge graphs: individual citizens that want to have control over their own data, with applications ranging from personal archiving all the way up to a personal digital assistant. The book is a collection of accessible contributions that open the door to this new vision on personal knowledge management.
Frank van Harmelen 教授，阿姆斯特丹自由大学知识表示教授
Prof Frank van Harmelen, Professor of Knowledge Representation at the Vrije Universiteit Amsterdam
作为一名生产力教练，我主要使用 PKG 来使知识可付诸实践，这对知识的吸收、发展和输出都有影响。Velitchkov 和 Anadiotis 在《个人知识图谱》中收集的文章涵盖了各种重要的 PKG 主题。有些文章更具哲学性，有些文章更实用，但所有这些都加深了我对如何充分利用我使用的 PKG 工具的理解。
As a productivity coach, I use PKGs primarily for making knowledge actionable—which has implications for the intake, development, and output of knowledge. The essays Velitchkov and Anadiotis have assembled in Personal Knowledge Graphs cover a wide variety of important PKG topics. Some essays are more philosophical, some are more pragmatic, but all of them deepened my understanding of how I can get the most out of the PKG tools I use.
RJ Nestor，生产力和思维工具专家
R.J. Nestor, Productivity and Tools for Thought expert
随着个人需要管理的数据变得越来越复杂，用于协助这一过程的工具和实践也越来越多。虽然新一代工具不一定基于开放和企业图谱方法，但似乎在某种程度上与它们趋同。
As the data that individuals need to manage is becoming increasingly complex, there has been a rise in the development of tools and practices to assist in this process. This new generation of tools, although not necessarily based on open and enterprise graph approaches, seem to be converging with them on some level.
这些工具允许个人以个人知识图谱的形式来管理他们的数据，通过可以遍历链接内容的边缘进行交互体验，类似于与“思考伙伴”一起探索的方式。
These tools allow individuals to manage their data as personal knowledge graphs, experienced interactively with edges that can be traversed linking content, in a manner akin to explorations with a “thinking partner”.
这本及时的书彻底回顾了围绕个人知识图谱的当前研究，旨在增强个人用户的能力，提高生产力、数据素养、主权和互操作性，并强调未来的方向。
This timely book thoroughly reviews current research around personal knowledge graphs, with the aim to empower individual users, promoting productivity, data literacy, sovereignty, and interoperability, as well as highlighting future directions.
伦敦大学金史密斯学院 TCIDA（钨智能数据分析中心）主任 J. Mark Bishop 教授
Prof J. Mark Bishop, Director, TCIDA (Tungsten Centre for Intelligent Data Analytics), Goldsmiths, University of London
如今，我们常常为了一点小事和一点安全感而放弃个人隐私。但随着我们的软件和数据挖掘系统变得越来越强大，我们必须重新掌控我们的数据和生活，而个人知识图谱有可能在为时已晚之前为我们做到这一点。
Too often today we give away our personal privacy for trinkets and for a little bit of security. But as our software and data mining systems get more powerful it’s crucial that we seize back control over our data and our lives and personal knowledge graphs have the potential to do that for us before it’s too late.
Dan Jeffries，人工智能基础设施联盟董事总经理、作家、未来学家、工程师和系统架构师
Dan Jeffries, Managing Director, AI Infrastructure Alliance, Author, Futurist, Engineer, and Systems Architect
新一代笔记工具可帮助我们快速将想法整理成知识图谱。与使用孤立的工具进行笔记相比，集成个人知识图谱将使用户能够创建更有价值、更有用的知识。
A new generation of note-taking tools helps us quickly organize thoughts as knowledge graphs. Integrating Personal Knowledge Graphs will allow users to create more valuable and helpful knowledge than if note-taking is done with siloed tools.
本书由一些世界顶尖的知识图谱专家撰写，致力于探索和突破个人知识图谱的界限。
This book, written by some of the world’s leading experts on knowledge graphs, promises to explore and push the boundaries of Personal Knowledge Graphs.
Dan McCreary，Optum 人工智能和图形领域杰出工程师
Dan McCreary, Distinguished Engineer in AI and Graph at Optum
个人知识图谱
Personal Knowledge Graphs
互联思维可提高生产力、创造力和探索能力
Connected thinking to boost productivity, creativity and discovery
編輯
Edited by
伊沃·维利奇科夫
Ivo Velitchkov
編輯
Edited by
乔治·阿纳迪奥蒂斯
George Anadiotis
2023 年由 Exapt Press 首次出版
First published in 2023 by Exapt Press
版权所有 © 2023 Exapt Press，Ashleigh Faith
Copyright © 2023 Exapt Press, Ashleigh Faith
编辑：Exapt Press 主编 Rob Worth
Edited by Rob Worth, Editor-in-Chief, Exapt Press
封面设计：Andrew Brown，DesignForWriters.com
Cover design by Andrew Brown, DesignForWriters.com
作者的精神权利已得到维护。
The moral right of the authors has been asserted.
保留所有权利。未经出版商书面许可，不得以任何形式或任何电子、机械或其他已知或后来发明的方式，包括影印和录制，或在任何信息存储或检索系统中重印、复制或使用本书的任何部分。
All rights reserved. No part of this book may be reprinted or reproduced or used in any form or by any electronic, mechanical, or other means, now known or hearafter invented, including photocopying and recording, or in any information storage or retrieval systems, without permission in writing from the publishers.
本书网站：https://PersonalKnowledgeGraphs.com
Book website: https://PersonalKnowledgeGraphs.com
建议参考：
Suggested reference:
Velitchkov, I. & Anadiotis, G. (Eds.) (2023)。个人知识图谱：连接思维以提高生产力、创造力和发现能力。Exapt Press。
Velitchkov, I. & Anadiotis, G. (Eds.) (2023). Personal Knowledge Graphs: Connected thinking to boost productivity, creativity and discovery. Exapt Press.
您可以在英国图书馆获取该书的 CIP 目录。
A CIP catalogue for this book is available from the British Library.
ISBN：978-1-914549-08-3（平装本）
ISBN: 978-1-914549-08-3 (paperback)
ISBN：978-1-914549-09-0（电子书）
ISBN: 978-1-914549-09-0 (eBook)
内容
Contents
Ashleigh Faith 博士
Dr. Ashleigh Faith
Past, present, and future of Personal Knowledge Graphs
1. 个人知识图谱——为什么、是什么以及在哪里？
1. Personal Knowledge Graphs – Why, what, and where to?
伊沃·维利奇科夫
Ivo Velitchkov
2. Niklas Luhmann 的个人知识图谱
2. The Personal Knowledge Graph of Niklas Luhmann
伊沃·维利奇科夫
Ivo Velitchkov
3. 评估个人知识图谱工具的框架
3. A framework for evaluating Personal Knowledge Graph tools
奥梅斯·巴尔特斯和乔治·阿纳迪奥蒂斯
Omes Baltes and George Anadiotis
Concepts, practices and visions
4. 知识的非故意部分在知识图谱中的决定性作用
4. The decisive role of the unintentional part of knowledge in PKGs
法布里斯·加莱
Fabrice Gallet
5. 思维算法，是应对 PKG 复杂性的必备 GPS？
5. Algorithms of thought, the must-have GPS for navigating the complexity of PKGs?
法布里斯·加莱
Fabrice Gallet
6. 利用 SEN 将桌面扩展为个人知识图谱
6. Extending the Desktop into a Personal Knowledge Graph with SEN
格雷戈尔·罗森瑙尔
Gregor Rosenauer
Use cases, prototypes, and implementations
7. 图表不是关键，而是引导我们找到关键
7. Graphs aren’t the thing, they’re the thing that gets us to the thing
马蒂纳斯·尤塞维丘斯
Martynas Jusevičius
8. 利用语义网标准构建个人知识图谱
8. Leveraging Semantic Web Standards for Personal Knowledge Graphs
Maribel Acosta 和 Omes Baltes
Maribel Acosta and Omes Baltes
9.图像、个人知识和多模态图
9. Images, Personal Knowledge and Multi-Modal Graphs
玛格丽特·沃伦
Margaret Warren
10. Agora 是一个社交知识图谱
10. The Agora is a Social Knowledge Graph
爱德华多·伊万内茨 (弗朗西安)
Eduardo Ivanec (Flancian)
前言
Foreword
Ashleigh Faith 博士
Dr. Ashleigh Faith
俗话说，智慧建立在知识的基础上，道路由经验铺就。我们现在已经掌握了知识部分。知识图谱已成为主流，并用于日常应用中。每天都会产生前所未有的大量数据，但尽管数据如此庞大，由于过去知识载体的技术衰退，我们失去的数据比获得的数据还多。
It is said that wisdom is built on foundations of knowledge and a road is paved by experience. We have the knowledge part down by now. Knowledge graphs have gone mainstream and are used in everyday applications. An unprecedented amount of data is being created every day but despite this deluge, we lose more data than we gain because of the technological decay of past knowledge vehicles.
我们与地点、人物和事物的数字副本共存。模式匹配和语言模型正在学习数十亿（并且还在增加）人类对话的怪癖，过去与现在的来回交织，人类对话的跳跃效应。人类精神和独创性的表象现在被写成散文、艺术和数字副本，复活了那些先行者并思考未来。道德和界限正在被划定，尽管可能太慢而跟不上，这些想法正在被探索、抛弃并重新拾起。
We live alongside digital copies of places, people, and things. Pattern matching and language models are learning billions (and counting) of the quirks of human dialog, the back-and-forth weave of the past with the current, a leap-frog effect of human conversation. Semblances of the human spirit and ingenuity are now composed into prose, art, and digital copies, resurrecting those who have come before and casting thought to what will come. The ethics and boundaries are being drawn, albeit likely much too slow to keep up, and the ideas are being explored, abandoned, and picked back up again.
但是，如果我们希望从所有这些知识中获得智慧，就缺少一个关键要素。那就是经验。通往智慧的道路是由知识构建的，由经验铺就的。现在，个人知识图谱 (PKG) 为智慧增色不少。
But there is a critical component missing if we wish to attain wisdom from all this knowledge. It is experience. The road to wisdom is built from knowledge and paved by experience. Now, supercharged with Personal Knowledge Graphs (PKGs).
个人经历绘制了每个人积累的知识图谱。经验是实验、失败、突破（无论大小）、重新组合、改进和逆转、重新配置以及在过去和未来的基础上构建的循环，所有这些都是为了迈出下一步。从这些经验中解读和学习可以产生经验教训。分享、传递和借鉴他人的经验教训可以创造智慧。现在，个人知识可以结构化，增加意义和见解，可见且可操作，由您和您的家人拥有和操作，随时可以让您深入了解您拥有但可能没有意识到的智慧深度。
Personal experiences create a map of each person’s accumulated knowledge. Experience is the cycle of experiments, failures, breakthroughs (small and large alike), remixes, revamps and reversals, reconfigurations, and building on what was and what is to come, all to take the next step, and the next. Interpreting and learning from those experiences creates lessons. Lessons shared, passed, and built upon by others create wisdom. And now, personal knowledge can be structured, adding meaning and insights, visible and actionable, owned and operated by you and yours, ready for you to peer into the depths of wisdom that you have, but perhaps were not aware of.
正如伊沃·维利奇科夫 (Ivo Velitchkov) 在其章节中所探讨的那样，打破创新障碍并找到新的探索途径、未曾探究的石头、未曾吸取的教训只是 PKG 提供一种方法来理解我们所有人的经验结构的几个例子。
As Ivo Velitchkov explores in his chapter, breaking down the barriers to innovation and finding new pathways to explore, stones unturned, lessons unmarked are just a few examples where PKGs provide a way to make sense out of the fabric of experiences we all have.
有这样一个关于谷歌过去的故事，据说有一个项目不仅要向谷歌传授知识，还要传授智慧。谷歌知识图谱出现了，但智慧部分却没有出现，或者至少还没有出现。你看，捕获网络数据可以为你提供知识、数据点以及它们与其他事物的关系，但不能告诉你为什么它们对个人或特定情况很重要。这种额外的解释需要从一个人的经历和环境的角度来传递知识，以及一种捕获和整理知识的方法，这就是知识图谱工具发挥重要作用的地方，也是 George Anadiotis 章节的重点。
There is a tale of Googles-past where there was rumored to be a project to impart, not only knowledge to Google, but also wisdom. The Google Knowledge Graph came to be, but the wisdom part did not, or at least not yet. You see, capturing the web’s data gives you the knowledge, the data points, and how they relate to other things, but not why those matter to individuals or specific situations. That additional interpretation requires the lens of a person’s experiences and circumstances for the knowledge to be passed through, and a way to capture and codify it, which is where tools for PKGs play an important role and which is the focus of George Anadiotis’s chapter.
知识和经验的编织依赖于解读才能成为智慧，这就是为什么 PKG 能够让这些知识和经验成为一幅可供评估和解读的图画，从而为智慧之路增添活力。智慧本身无法被包含，只有提供可供解读的见解，而且是由特定的人根据自己的经验提供的。这种对个体的强调使得 PKG 不同于传统的知识图谱。在 PKG 中，最重要的节点是代表您作为个体的中心节点，而最重要的边是您解读您个人生活交响曲音符的方式、构成您世界的编织方式以及您如何解读它。
The weave of knowledge and experience depends on interpretation to become wisdom, which is why PKGs, which allow these to become a picture to be assessed and interpreted, supercharge the road to wisdom. Wisdom cannot be contained per se, only insights offered for interpretation, and that by a specific person based on their own experiences. This emphasis on the individual is what makes PKGs different from a traditional knowledge graph. In a PKG, the most important node is the center node representing you as an individual, and the most important edges are the way you interpret the notes to your own personal symphony of life, the weave that makes your world, and how you interpret it.
每个人都有自己的生活，并与其他人的生活互动，创造出独特的经验教训和知识线索，编织成智慧的被子，我们也可以将其传递给下一代。安慰。成长。守护。激励。历史上最常用的载体是故事，这些叙述将知识和经验汇集在一起，将智慧传递给下一代。但故事往往会扭曲、改变，呈现出自己的生命，不幸的是，它们也会随着岁月的流逝而消失。文化遗产，那些让你成为你的故事，你的文化成为你日常生活的调味品，你的反应和感受，传统和语言，都包含在口头故事或通常过时的格式（如磁带或纸张）中，现在可以获得第二次生命。这些可以因地区和家庭而异，在不同文化中分享主题，并且随着年龄的增长而根据当代人的需求而变化。加拿大、爱尔兰、肯尼亚和其他文化遗产和档案组织已经开始探索使用图表来保存这些叙述。
Each person has lived a life, and interacted with other lives, creating a unique weft of lessons and threads of knowledge that weave into a quilt of wisdom that we can also wrap around the next generation. To comfort. To grow. To guard. To inspire. The most used vehicle for this throughout history has been stories, the narratives to pull together the knowledge and experiences to pass on wisdom to the next. But stories tend to warp, change, take on a life of their own, and unfortunately, they also get lost down the years. Cultural heritage, the stories that make you you, your culture a spice to your everyday life, the way you react and feel, traditions and language, all contained in verbal stories or often outdated formats like tape or paper can now get a second life. These can change by region and family, share themes across cultures, and have versions across the years that change according to the needs of the current generation. Efforts in Canada, Ireland, Kenya, and other cultural heritage and archival organizations have started exploring the use of graphs for saving these narratives.
正如玛格丽特·沃伦在其关于图像在个人知识中所扮演的角色的章节中所探讨的那样，PKG 提供了一种方法来包含这些故事的交织、节点以及不同叙事之间的起伏，所有这些都构成了我们许多人如何定义自己、我们来自哪里以及我们要去哪里的支柱。比较 PKG 以寻找见解和意义不仅仅是学者的事情，也是在日常生活中寻找生活细微差别背后的意义的美妙之处。
As Margret Warren explores in her chapter focused on the role images play in someone’s personal knowledge, PKGs offer a way to contain the interweaving of these stories, the nodes, and the ebbs and flows between different narratives, all forming the backbone of how many of us define ourselves, where we came from, and where we are building to go. The comparison of PKGs to find insights and meaning is not the stuff of scholars alone, but the beauty of the everyday in finding meaning behind the nuances of your life.
正如法布里斯·加莱和爱德华多·伊万内茨在各自的章节中所探讨的那样，跨生活、跨情境、跨代际、跨项目、跨同事、跨文化和跨学科地分享知识和经验，从一种世界观与另一种世界观的相互作用中获得真知灼见，所有这些都为智慧的构建和共享创造了更大的机会，而这些个人联系的建立往往是一个丰富的研究领域，研究意向性和一个人与知识的联系以及他们所编织的个人智慧，以及从这种共享交织中获得的好处。
As Fabrice Gallet and Eduardo Ivanec explore in their respective chapters, the sharing of knowledge and experiences across lives, situations, generations, projects, colleagues, cultures, and disciplines, deriving insights from the interplay of one world view to another, all build to a greater opportunity for wisdom building and sharing, and the creation of these personal connections is often a rich area of study on intentionality and a person’s connection to knowledge and their own personal wisdom they weave, and the good that can be derived from this shared interweaving.
经验及其解读也伴随着沉重的责任。经验、知识和智慧都是由每个人培养的。筛选出重要的东西、可以丢弃的东西或留到下次再用的东西，构建自己的思维殿堂是困难的，需要付出努力和时间。Fabrice Gallet 在他的章节中探讨了，在某种程度上，人们可能需要一套指南、指南或好的 GPS 来找到穿越洞察力、感知和知识迷宫的路，以发掘联系网络之下的智慧。你个人投资的很大一部分用于理解事物，而这些不能轻易拍卖或处理不当。智慧可能会被扭曲、曲解，用于从未打算或未得到充分理解的目的。
Experiences and their interpretation come with a heavy responsibility as well. Experiences, knowledge, and wisdom all are cultivated by each individual. Sifting through what matters, what can be discarded, or tucked away for another day, building your own mind palace is hard and takes work and time. Fabrice Gallet explores in his chapter that in a way, one might need a set of directions, a guide, or good GPS to find their way through the maze of insights, perceptions, and knowledge to unearth the wisdom beneath the web of connections. A large portion of your personal investment goes into making sense of things and these cannot be so easily auctioned off or mishandled. Wisdom can be misshapen, misinterpreted, used for purposes never intended or well understood.
“我的生活，我的数据”这一口号已成为数据访问权越来越广泛的一面旗帜，个人经验受到管理数据收集、访问和使用的法律法规（如 GDPR）的保护。经验和智慧是生活中非常个人化的产物，而且往往是健康快乐生活的基础。分享您的经验和智慧来帮助他人是一种珍贵的礼物。添加他人的经验和智慧可以积累、发展成新的东西，并添加零碎的东西来帮助解决各种情况的长尾和短尾问题，从而更专业地报道任何给定的主题。但这就是为什么信任和所有权是 PKG 如此重要的支柱。在天空中写下您个人或他人奋斗的耀眼信息会带来混乱。将石头扔进平静的池塘只是为了看看涟漪，测试您能让水变得多么湍急，或者欺骗他人试水，都意味着不信任。两者都没有建立稳定的基础，通往智慧的道路充满曲折和坎坷，如果不保持和改善信任和所有权，我们可能看不到尽头。PKG 将所有权赋予创建数据的个人。毕竟，数据是他们自己的经验和解释，因此如果存在虚假，至少也只是一种自私的欺骗。这些主题已在语义网标准中进行了探讨，这也是 Omes Baltes 和 Maribel Acosta 章节的主题。
The mantra “my life, my data” has been a flag raised in light of greater access to data, and personal experience is covered under the laws and regulations that govern data gathering, access, and use, such as GDPR. Experiences and wisdom are deeply personal artifacts to life and often are foundations of a healthy and happy life at that. Sharing your experiences and wisdom to help others is a precious gift. Adding the experiences and wisdom of others builds up, grows into something new, and adds odds and ends to help the long and short tails of situations for more specialized coverage of any given topic. But this is why trust and ownership are such critical pillars of PKGs. Writing a blazing message across the sky of your personal struggles, or those of others spells havoc. Throwing a stone into a calm pond just to see the ripples, testing how turbulent you can make the waters, or fooling others into testing those waters, spells distrust. Neither creates a stable foundation to build on, making for a twisted and bumpy road to wisdom and one we may not be able to see the end of if trust and ownership are not maintained and improved. PKGs put ownership on the individuals creating the data. The data is, after all, their own experiences and interpretations, so if there is falsehood, it is at least only a self-serving deception self-contained. These topics have been explored in standards for the semantic web, which is also the theme for Omes Baltes and Maribel Acosta’s chapter.
当 PKG 被共享或 PKG 用于健康和身份等关键应用时，需要验证、制衡、匿名和治理来保持安全平衡。这看起来如何还有待确定，但与任何关键时刻一样，信任很容易失去，很难获得。利用过去的关键时刻的智慧，我们可以学到如何测试我们与谁分享什么以及为什么分享，并根据我们的信任因素进行衡量。
When PKGs are shared or where a PKG is used for critical applications such as health and identity, this is where verification, checks and balances, anonymity, and governance are needed to maintain an equilibrium of safety. What that looks like is still to be determined, but as with any pivotal moment, trust is easily lost and hard to gain. Using the wisdom of past pivots provides us the learnings to test what we share with who and why, and measure it against our trust factors.
智慧不仅与你过去和当前情况的解读有关，还与对未来决策的预测有关。正如 Martynas Jusevicius 一章所探讨的那样，个人知识图谱创建了一个窗口，但这并不是主要吸引力。吸引力在于个人知识图谱帮助我们做什么、了解什么和被了解什么。毫无保留地了解你拥有什么、它来自哪里、你是如何获得或学习某些东西的、谁有访问权限或相同的知识或经验、你可以与谁分享这些以加速了解情况，以及这些如何随着时间的推移和不同的干预而发生变化，这些都是做出更明智决策的解读。看到这些经验的交织可以识别漏洞或焦点、成长和探索的空间，以及你自己的数据或其他贡献或解释你图表数据的人的数据中存在错误的地方。经验经常与他人分享以获得见解和理解。想象一下与你的财务顾问分享你的个人知识图谱，这样他们就可以直接指导你，寻求更平衡的搜索体验，或者规划你的癌症幸存者之旅，包括治疗、药物、提供者和支持，与他人分享。能够分享这些事情可以增强你的决心和信心，让你能够指导或支持他人，或帮助解决争议或不正确的数据和解释。虽然这些例子更多的是承诺而不是现实，但 PKG 的现状（其中许多都可以在本书中探讨）正在为实现未来的创新和社区建设奠定基础。
Wisdom is not only connected to the interpretations of your past and current situations but also predictions for future decisions. As explored in Martynas Jusevicius chapter, personal knowledge graphs create a window but that is not the main attraction. The appeal is what PKGs help us to do, to know and be known. Knowing without blinders what you have, where it came from, how you obtained or learned something, who has access or the same knowledge or experiences, who can you share that with to gain an accelerated view on a situation, and how these change over time and with different interventions, these all are interpretations to make more informed decisions. Seeing the weave of those experiences can identify holes or focal points, space to grow and explore, where there are errors in your own data or that of others who contribute to or interpret data from your graph. Experiences are often shared with others to gain insights and understanding. Imagine sharing your PKG with your financial advisor so they can guide you straight, seeking a more balanced search experience, or mapping out your cancer survivor journey with treatments, drugs, providers, and support to share with others. Being able to share these things can strengthen your resolve and confidence with others who can guide or support, or help to resolve disputes or incorrect data and interpretations. While these examples are more the promise than reality, the here and now of PKGs, many of which can be explored throughout this book, are setting the groundwork to bring the future innovations and community building to reality.
对于使用 PKG 数据设计和构建体验的人来说，分享和发现隐藏的联系尤其有趣。实际上并不存在普遍的智慧。每个人都是独一无二的，因此虽然他们的解释可能相似，但对他们具体情况的解释都是独一无二的。那些分享或创建个人数据 PKG 的人可以借助 PKG 提供比以往更细致、更亲密的体验，这些体验基于类似的模式，带来欢乐和惊喜、愉悦和兴奋，但正如 Gregor Rosenauer 的章节所探讨的那样，PKG 尚未成为主流，因此它们的数据可能很稀疏。我的祖母或我五岁的侄女还不知道如何制作 PKG。他们都依赖口头和书面分享，正如 Rosenauer 在桌面上对 PKG 的描述所见，而不是常见的 PKG 的动态表示。因此，虽然 PKG 前景广阔，但仍有机会帮助普通民众进入 PKG 领域并挖掘可能隐藏的智慧。
Sharing and discovering hidden connections is especially interesting to explore for those using PKG data to design and build experiences. There is no such thing as universal wisdom, not really. Each person is unique and therefore while their interpretations may be similar, the interpretations of their specific situations are all unique to them. Those sharing or creating PKGs from personal data can offer experiences that are more granular and intimate than ever before with the help of PKGs, based on similar patterns to bring joy and surprise, delightment and excitement, but as Gregor Rosenauer’s chapter explores, PKGs are not yet mainstream, and therefore their data can be sparse. My grandma or my five-year-old niece do not yet have the know-how to make a PKG. They both fall back on verbal and written sharing, as can be seen in Rosenauer’s depiction of a PKG on a desktop, and not the dynamic representations of PKGs that are so common. So, while PKGs have great promise, there is still opportunity to help everyday folk get into the PKG space and tap the wisdom that might be hidden.
踏上新的道路既可怕又美丽，包含着可能性和机遇，但也存在不确定性和未兑现的承诺。通往智慧的道路也不例外。许多人声称“下一件大事”。我也不是说个人知识图谱就是下一件大事。但它们是未来承诺的下一件大事的基础。PKG 不是“事物”，但它们是“事物”的一种手段，如果我们愿意的话，它们可以帮助铺平通往智慧的道路，无论是个人还是集体。这本书汇集了多年来一直在个人知识领域探索的作者，他们都是为了解开盒子里的东西，或者至少是可能性。我们会发现什么？在接下来的几章中，加入我们，探索当今通往智慧的道路是如何铺设的，它来自哪里，有哪些未解决的问题和未解决的承诺，以及 PKG 可以作为一种机制来构建你的知识块，为你铺平通往智慧的道路，并希望在此过程中传授和分享一些知识。
New roads traveled are scary and beautiful things, containing possibilities and opportunity, but also uncertainty and the potential for promises unkept. The road to wisdom is no different. There have been many claims to “the next big thing.” And I am not saying personal knowledge graphs are the next big thing either. But they are the foundations for the next big things that are the promises to come. PKGs are not the “thing,” but they are a means to the “thing” that can help pave the road to wisdom, personally and collectively, if we so choose. This book is a compilation of authors who have been journeying in Personal Knowledge for years, all to unravel what is in the box, or at least the possibilities. What will we find? Join us in the next few chapters in exploring how the road to wisdom is being paved today, where it has come from, the open questions and open promises, and where PKGs can serve as a mechanism to build your knowledge blocks and pave your path to wisdom and to hopefully impart and share some along the way.
祝大家旅途顺利，享受旅途吧！
Fair travels all, and enjoy the ride!
Ashleigh Faith 博士
Dr. Ashleigh Faith
知识图谱领域的研究员
Researcher in the Knowledge Graph space
Isa DataThing 教育 YouTube 频道 @AshleighFaith 的创始人
Founder of Isa DataThing Educational YouTube Channel @AshleighFaith
介绍
Introduction
边缘跨越事物之间的缝隙不是为了填充它们而是为了穿越空间，侦察固体物体的形状并画出新的线条。
Edges cross gaps between things not in order to fill them but in order to traverse the space, to reconnoiter the shapes of solid things and draw new lines.
《好奇的头脑：连接的力量》，Zurn 和 Bassett
Curious Minds: The Power of Connection, Zurn and Bassett
好奇心是什么？你很可能很好奇这本书讲的是什么，但现在你可能想知道好奇心本身与它有什么关系。结果发现，两者关系很大。Zurn 和 Bassett (2022) 展示了好奇心不是填补知识空白，而是建立联系。它是一种边缘工作。它是图形形状的。而且，事实上，信息通常是图形形状的，或者当它最初以不同的形状出现时必须是图形形状的，而我们需要一个图形来整合它。
What is curiosity? Most likely, you are curious to learn what this book is about, but now you may wonder what curiosity itself has to do with it. A lot, it turns out. Zurn and Bassett (2022) show how curiosity is not about filling knowledge gaps but about making connections. It’s an edgework. It’s graph-shaped. And, as it happens, information, in general, is graph-shaped or has to be when it initially comes in different shapes, and we need a single one to consolidate it.
欢迎阅读第一本关于个人知识图谱的书！现在还为时过早，也许我们对 PKG 还不够了解，但我们确信这“是一个东西”，而且是一个有价值的东西。我们，那些在这个我们同意称之为 PKG 的广场上相聚的人，沿着不同的道路前来，寻找不同问题的解决方案。有些人来这里是为了提高他们的生产力，有些人来这里是为了支持他们的创造力，而其他人则在寻求重新控制他们的数据。
Welcome to the first book on Personal Knowledge Graphs! It is still early, and maybe we don’t know enough about PKG yet, but we are certain that this “is a thing” and a valuable one. And we, those that met at this square we agreed to call PKG, came along different streets, searching for solutions to different problems. Some came looking to boost their productivity, some to support their creativity, yet others were on a quest to take back control over their data.
肖恩·加拉格尔 (Shaun Gallagher) (2022) 在谈到胡塞尔的思想时写道：“知识可以理解为由推理关系联系在一起的相互关联的命题系统”。因此，“知识图谱”只是一个同义反复。知识就是一张图。由于人是那些能够知道的人，我们在创造这个术语之前就一直在使用个人知识图谱。
Knowledge “can be understood as a system of interconnected propositions linked by inferential relations”, wrote Shaun Gallagher (2022) about the ideas of Husserl. “Knowledge graph” then is simply a tautology. Knowledge is a graph. And since persons are those who can know, we had been using personal knowledge graphs way before we coined the term.
如今，“知识图谱”是指仅使用节点和边来表示知识的一种方式，其中节点表示感兴趣的实体，边表示这些实体之间的关系。这种方法的好处使其传播开来，首先是开放和企业知识图谱，现在则是个人知识图谱。有些好处是所有类型的知识图谱都具有的。知识图谱可以吸收复杂性。它们灵活且善于处理变化。变化可以是任何类型的变化，环境的变化或需求和偏好的内部变化。这是知识图谱的特殊功能，技术上称为“后期绑定”，或者如 Dave McComb 所说，“后期模式”。换句话说，当有足够的知识来做决定时，您可以随时添加或扩展模式。所有类型的知识图谱在空间上都具有这种灵活性，以便它们可以统一多样性，在时间上也可以适应变化。但个人知识图谱还有另一个特殊功能。它们是惊喜的生成器。它们可以根据需求提供意外发现。
What nowadays is meant by “knowledge graph” is a way of representing knowledge by using only nodes and edges, where nodes represent entities of interest and edges represent relationships between these entities. The benefits of this approach made it spread, first for open and enterprise knowledge graphs, and now for personal knowledge graphs. Some of the benefits are common for all types. Knowledge graphs can absorb complexity. They are flexible and good at handling change. It can be a change of any kind, a change of the environment or an internal change of needs and preferences. This is the special feature of knowledge graphs, technically called “late binding” or, as Dave McComb put it, “schema late”. In other words, you add or extend the schema as you go and when needed, when there is enough knowledge to decide. All kinds of knowledge graphs share this flexibility in space so that they can unify diversity and in time so that they can accommodate change. But personal knowledge graphs have another special feature. They are generators of surprise. They can deliver serendipity on demand.
塞内加和马可·奥勒留都记过某种日记，这可能是“普通书籍”的最早例子——这是管理个人知识的第一种技术。在启蒙运动时期，当约翰·洛克写出一种组织普通书籍的先进方法时，普通书籍得到了传播和改进，并找到了新的使用方式。这种方法广为流传。它不仅在塞内加的时代被使用，而且在很久以后也被查尔斯·达尔文等思想家使用。但也许第一个朝着灵活性方向发展的技术变革是托马斯·哈里森在 1740 年的发明。他的“研究方舟”是一个木制柜子，文件卡片挂在锡板上。这种开放，加上更复杂的索引系统的开发，为古老的摘录艺术带来了进一步的创新和功能转变，正如阿尔贝托·塞沃里尼所说，“从记忆辅助工具到二级记忆”。然而，直到 Niklas Luhmann 的 Zettelkasten 出现，高级交叉引用、延迟分类和建立意外联系才达到我们现在拥有更先进技术的个人知识图谱所期望的水平。一段时间以来，知识图谱和个人知识管理并行发展，但最近这种情况发生了变化。起初，只有少数基于图谱的个人知识工具，然后在几年内，当一个真实的工具向我们展示世界是一个图谱时，它们像病毒一样传播开来，而这个图谱比我们想象的更加紧密地相互联系。
Seneca and Marcus Aurelius kept a certain kind of journal, which may be the earliest examples of “commonplace books” – the first technology for managing personal knowledge. Commonplace books spread and improved and found new ways of being used during the Enlightenment when John Locke wrote an advanced method for organising them. This method spread far and wide. It was used not only during his time but much later by thinkers such as Charles Darwin. But maybe the first technological change towards flexibility was the invention of Thomas Harrison in 1740. His “Ark of Studies” was a wooden cabinet in which file cards were hooked on tin plates. This opening up, together with developing more sophisticated systems of indexing, brought further innovations to the ancient art of excerption and a functional shift, as Alberto Cevolini put it, “from memory aids to secondary memories”. Yet it wasn’t until Niklas Luhmann’s Zettelkasten, when advanced cross-referencing, delayed classification and making unexpected connections reached the level that we now, armed with more advanced technologies, expect from our personal knowledge graphs. For some time, knowledge graphs and personal knowledge management developed in parallel, independently, but recently this has changed. At first, there were only a few graph-based tools for personal knowledge, and then they spread like a virus in the years when a real one showed us that the world is a graph, interlinked more densely than we thought.
现在，在 2023 年，您手中有第一本关于个人知识图谱的书。像任何其他书一样，这是一段旅程。或者更确切地说，是许多旅程。对于我们 Ivo 和 George 来说，这是一次个人旅程。开始它并不需要太多勇气，我们觉得我们必须这样做，但坚持下去并完成它需要一些勇气。[¹] 对于所有其他加入我们的贡献者——Eduardo、Fabrice、Gregor、Margaret、Maribel、Martynas 和 Omes——来说，这是一次旅程。对于您，读者来说，这将是一次旅程。我们希望这将是一次穿越新地方的愉快旅程，有时可能会颠簸不已，但希望在沿途的某些转折处，它会揭示出您希望进一步探索的迷人风景。并且根据您是谁以及您在寻找什么，您可能会发现不同的有趣事物。这本书应该是一本适合任何好奇心的人的书。无论您做什么，尤其是涉及处理大量信息的工作，了解图表如何提供帮助都是值得的。您可能希望提高工作效率，更好地管理研究，组织知识，以激发创造力并定期为您带来意想不到的联系，或者您可能只是需要一种更好的方式来做笔记或管理项目和数字收藏。如果您是工具制造者，它可能会为您正在进行的工作带来新的想法和见解，或激励您创造全新的东西。这本书不仅旨在通过它带来的东西激发人们的兴趣，还旨在通过它没有带来的东西激发人们的兴趣，即要填补的已知和未知的空白，以及要探索和连接的新领域。
And now, in 2023, you have in your hands the first book on personal knowledge graphs. Like any other book, it is a journey. Or rather, many journeys. It’s a personal journey for us, Ivo and George. It didn’t take much courage to start it, we felt we had to, but it took some to keep going and finish it. [ ¹ ] It was a journey for all the other contributors – Eduardo, Fabrice, Gregor, Margaret, Maribel, Martynas and Omes – who joined us. And it will be a journey for you, the reader. We hope it will be a nice walk through new places, maybe a bumpy ride at times, but hopefully, along the way, at some turns, it will reveal alluring landscapes that you would wish to explore further. And depending on who you are and what you are looking for, you may find different things of interest. This is supposed to be a book for any curious mind. Whatever you do, but especially if it involves working with a lot of information, it’s worth knowing how graphs can help. It may be that you’d like to improve your productivity, manage your research better, organise your knowledge so that it stimulates your creativity and regularly brings you surprising connections, or you may simply need a better way to take notes or manage your projects and digital collections. If you are a toolmaker, it might bring new ideas and insights on what you are already working on or inspire you to create something entirely new. And this book is meant to provoke not only by what it brings but by what it doesn’t, the known and unknown gaps to be filled, and the new territories to be explored and connected.
笔记
Notes
[1] 请参阅https://personalknowledgegraphs.com/#/page/heroes%20journey%3A%20notes%20towards%20a%20pkg%20book
[1] See https://personalknowledgegraphs.com/#/page/heroes%20journey%3A%20notes%20towards%20a%20pkg%20book
个人知识图谱的过去、现在和未来
Past, present, and future of Personal Knowledge Graphs
第 1 章
Chapter 1
个人知识图谱——为什么、是什么以及去哪里？
Personal Knowledge Graphs – Why, what, and where to?
伊沃·维利奇科夫
介绍
Introduction
知识图谱的采用正在迅速增长。从开放数据发布到企业数据集成再到机器学习，知识图谱的三个特性使其在各种用例中都具有吸引力。第一个特性是知识图谱可以成为异构数据结构之上的通用抽象层，因此可以统一来自独立数据源的数据。其次，在语义知识图谱中，数据、含义和规则共存于图中，使它们具有自描述性，并且可能独立于使用它们的应用程序。第三，知识图谱提供了灵活性，可以帮助人们处理复杂问题，同时以极低的成本适应变化。
The uptake of knowledge graphs is growing rapidly. Three of their qualities make them attractive in various use cases, from open-data publishing through enterprise data integration to machine learning. The first quality is that knowledge graphs can be a universal abstraction layer on top of heterogeneous data structures and so can unify data coming from independent data sources. Second, in semantic knowledge graphs, data, meaning, and rules live together in the graphs making them self-descriptive and potentially independent from applications that use them. And third, knowledge graphs provide flexibility that can help people deal with complex problems while also accommodating change at a very low cost.
那么，知识图谱进入个人信息和知识管理领域也就不足为奇了。除了开放和企业知识图谱之外，现在又出现了一个新事物：个人知识图谱 (PKG)。
Then it is not surprising to see knowledge graphs also enter the field of personal information and knowledge management. And apart from open and enterprise knowledge graphs, there is now a new kid in town: personal knowledge graphs (PKG).
但是，PKG 是什么？我们为什么要关心它？我们现在采用 PKG 的程度如何？我们用它们做什么？PKG 可能会如何发展？
But what are PKGs, and why should we care? Where are we now in adopting PKG, what do we use them for, and how is this likely to evolve?
本章试图回答这些问题和其他问题。第一部分阐明了 PKG 是什么，重点关注“知识”的概念。然后简要概述了 PKG 的第一批应用。第二部分转向认知科学，解释为什么 PKG 不仅仅是一种工具，而且是一个参与者。第三部分也是最大的部分提供了 PKG 演变的视角，并推断出未来可能的发展路径。
This chapter tries to answer these and other questions. The first section clarifies what a PKG is, with more attention to the concept of “knowledge.” Then there is a brief overview of the first applications of PKGs. The second section turns to cognitive science to explain why a PKG is not just a tool but also a participant. The third, and largest, section offers a view of the evolution of PKGs, extrapolating to a possible path in the future.
个人知识图谱
Personal Knowledge Graphs
[关联索引] 的基本思想是，任何项目都可以随意选择另一个项目。这是 memex 的基本功能。将两个项目绑定在一起的过程是最重要的。
[T]he basic idea of [associative indexing] which is a provision whereby any item may be caused at will to select immediately and automatically another. This is the essential feature of the memex. The process of tying two items together is the important thing.
万尼瓦尔·布什，《如我们所想》，1945 年
Vannevar Bush, As We May Think, 1945
什么是个人知识图谱？
What is a personal knowledge graph?
回答这个问题就是提出定义。这样做意味着定义是一件好事。然而“每个定义都有一个根本的弱点：它排除和限制”（Foerster & Poerksen，2002）。
Answering this question is an act of proposing a definition. Doing it implies that definitions are a good thing. And yet “every definition has a fundamental weakness: It excludes and limits” (Foerster & Poerksen, 2002).
关于定义
On definitions
定义并不存在于外界。它们是由人制定的。因此，它们具有个人性。即使达成一致，并且定义被许多人接受和使用，它也不会停止具有个人性，而只是变成了人际性。定义不仅由人制定，而且是为人而制定的。人们花费了大量精力来制定和商定一个定义，而不同的人有不同的兴趣和需求，对同一事物的不同定义将更好地为他们服务。
Definitions don’t exist out there. They are made by people. So they are personal. Even when an agreement is reached, and a definition is accepted and used by many, it doesn’t stop being personal, it just becomes inter-personal. Definitions are not only made by people but also for people. Much effort is spent carving and agreeing on one definition, while there are different people with different interests and needs and they will be better served by different definitions of the same thing.
阅读定义可以让你了解所定义的事物。[¹] 它让你了解一个单词或短语的含义，但不会自动获得更好的理解。为此，你需要了解不同的观点，在各种语境中看待该术语，最好是获得第一手经验。理解，即生动的定义，是经过一些互动后会出现的东西。这种互动可以是与 PKG 工具、关于 PKG 的书籍（例如你现在正在阅读的书籍）以及其他 PKG 用户和研究人员的互动。
Reading a definition lets you know about the thing defined. [ ¹ ] It gives you an idea of what is meant by a word or a phrase, but you don’t automatically get a better understanding. For that, you need to learn about different perspectives, see the term in a range of contexts, and, better still, get firsthand experience. The understanding, then, the living definition, is something that will emerge after some interaction. This interaction can be with PKG tools, books about PKGs, like the one you are reading now, and with other PKG users and researchers.
在使用知识图谱和本体制作词汇表时，定义既处于固定状态，又处于动态网络状态。用于创建同义词库的最流行的本体 [2] 是简单知识组织系统 (SKOS)。它有一个正式的词汇表，本身就是一个图，它为实例图、同义词库或分类法提供明确的语义。SKOS 有一个属性 [3] skos:definition，[4] 它将代表概念的节点链接到某种语言中人类可读的定义。[5] 但更重要的是，每个概念都通过语义关系（如更宽、更窄和相关）链接到其自身或外部词汇表中的其他概念。这样，除了互操作性增益之外，理解不仅来自于了解定义，还来自于了解所有这些关系。知识图谱是开放式的。添加新的边和节点不受预定义模式的限制。[6] 随着关系的增长，活生生的定义，即实际的理解，变得更加丰富。
When using knowledge graphs and ontologies for making glossaries, definitions appear in a fixed state and in a dynamic, networked state. The most popular ontology [2] for creating thesauri is the Simple Knowledge Organization System, SKOS. It has a formal vocabulary, itself a graph, that gives explicit semantics to an instance graph, a thesaurus, or a taxonomy. SKOS has a property [3] skos:definition, [4] that links a node representing a concept to a human-readable definition in a certain language. [5] But more importantly, each concept is linked to other concepts from its own or external vocabularies with semantic relations such as broader, narrower, and related. This way, on top of the interoperability gain, the understanding comes out of knowing not only the definition but also all these relations. Knowledge graphs are open-ended. Adding new edges and nodes is not restricted by a predefined schema. [6] And with the growth of relations, the living definition, the actual understanding, gets richer.
定义就像一个俱乐部，入口处有保镖检查想要进入的人是否符合资格。更正式地说，定义指定了必要条件和充分条件。我们可以在 OWL 本体中定义一个“个人知识图谱”类，并正式列出所有需要满足的条件（同样作为知识图谱的一部分），以便具有所需特征的资源可以归类为“个人知识图谱”。
Definitions are like a club with a bouncer at the entrance to check if those who want to enter are qualified. More formally, definitions specify necessary and sufficient conditions. We can define a class of “Personal Knowledge Graph” in an OWL ontology and formally list all the conditions, again as part of a knowledge graph, that need to be satisfied so that a resource having the required characteristics can be classified as a “Personal Knowledge Graph.”
总而言之，知识图谱本身有能力正式 [7] 定义任何事物，包括什么是 PKG。现在让我们用自然语言寻找 PKG 的非正式定义。
In summary, knowledge graphs themselves have the capabilities to formally [7] define anything, including what is a PKG. Let’s now look for an informal definition of PKG, in natural language.
既然我们需要定义个人知识图谱，那么我们首先需要知道什么是知识图谱。当知识图谱这个话题像病毒一样传播开来时，它们传播得越多，变体就越多，因此意见也千差万别。2019 年，Michael Bergman 在 1974 年以来出版的资料中发现了 27 种知识图谱的定义（Bergman，2019 年）。在这种分歧之后，随着《知识图谱》一书的出版（Hogan 等人，2022 年），出现了表面上的趋同。在那里，“知识图谱”被定义为：
Since we need to define personal knowledge graph, we need to know first what a knowledge graph is. When the topic of knowledge graphs went viral, just like viruses, the more they spread, the more variants appear, and so there was a great diversity of opinions. In 2019, Michael Bergman found 27 definitions for knowledge graphs (Bergman, 2019) in sources published since 1974. After this divergence, there was an ostensible convergence with the publication of the Knowledge Graphs book (Hogan et al., 2022). There, a “knowledge graph” is defined as:
用于积累和传达现实世界知识的数据图，其节点代表感兴趣的实体，其边代表这些实体之间的关系。
a graph of data intended to accumulate and convey knowledge of the real world, whose nodes represent entities of interest and whose edges represent relations between these entities.
这个定义很清晰，但依赖于对图和知识的先前的共同理解。
This definition is clear but relies on a prior common understanding of graph and knowledge.
图的概念比较简单。它在数学中被严格定义为一个有序三元组，由一组节点、一组边和一个将每条边映射到一对节点的函数组成。它的历史众所周知，始于 1736 年欧拉对柯尼斯堡七桥问题的负解。[8]
The concept of graph is easier. It is rigorously defined in mathematics as an ordered triple comprising a set of nodes, a set of edges, and a function mapping each edge to a pair of nodes. It has a well-known history starting with Euler’s negative resolution of the problem of the Seven Bridges of Königsberg in 1736. [8]
但是知识呢？它似乎被视为理所当然。上面的定义指出“旨在积累和传递知识”。但是知识可以存储和传输吗？要回答这个问题，我们应该更仔细地研究知识，不是带着定义知识的野心，而是阐明某些方面并重新聚焦注意力。这个术语被滥用，赋予知识它不可能拥有的特征，并掩盖了其他对一般和 PKG 来说都很重要的特征。
But what about knowledge? It seems to be taken for granted. The definition above states “intended to accumulate and convey knowledge.” But can knowledge be stored and transmitted? To answer that, we should get a closer look at knowledge, not with the ambition to define it, but rather to elucidate some aspects and refocus the attention. The term is abused in a way that endows knowledge with characteristics that it cannot possibly have and obscures others that are important both in general and for PKGs in particular.
知识
Knowledge
定义知识超出了本节的目的。相反，我将尝试提出一些观点，其中一个观点并不受欢迎，以改变目前 IT 和商业界普遍存在的理解。
To define knowledge is beyond the ambition of this section. Instead, I’ll try to bring a few perspectives, one of which unpopular, to shift the understanding from the one that currently dominates in both IT and business circles.
关于知识，尤其是在商业和信息背景下，有一种流行的说法是，知识比信息更有价值，而信息本身又比数据更有价值。该模型的支持者声称，数据本身毫无意义。例如，你看到 100，但你不知道它是年、美元还是其他什么。当你了解到它是度数的测量单位时，你仍然不知道它是关于角度还是温度的。即使你发现它是温度测量，除非你知道单位和温度是多少——空气、水或其他东西——否则它也无法提供太多信息。如果它是以摄氏度为单位的水测量，那么你可以将它与其他知识联系起来，并推断出水正在沸腾。听起来很有说服力，但它意味着信息只能参考数据来定义，知识只能参考信息来定义。至少在流行的数据-信息-知识-智慧 (DIKW) 金字塔中情况确实如此。更糟糕的是，它排除了唯一能够理解事物的人，即活着的个体。我在其他地方更详细地讨论过这些问题（Velitchkov，2017 年）。现在，让我们看看其他观点，这些观点虽然不那么巧妙，但可能错误性较小。
A popular narrative about knowledge, especially in the context of business and information, is that it is something more valuable than information, which itself is more valuable than data. Data, the supporters of that model claim, is meaningless on its own. For example, you see 100, but you don’t know if it is years or dollars or something else. When you learn it’s a measure of degrees, you still don’t know if it’s regarding angle or temperature. And even when you find out it’s a temperature measurement, that cannot bring much information unless you know the units and what is it the temperature of – air, water, or something else. If it is a water measurement in Celsius, then you can connect it with other knowledge and infer that the water is boiling. Sounds convincing and yet it means that information can only be defined in reference to data, and knowledge in reference to information. That is the case at least in the popular data-information-knowledge-wisdom (DIKW) pyramid. What’s worse, it excludes the one and only one who is capable of sense-making, the living individual. I have discussed these issues in more detail elsewhere (Velitchkov, 2017). Now, let’s look at other perspectives, which are not so neat, but are probably less wrong.
胡塞尔认为，“知识[…]可以理解为由推理关系联系在一起的相互关联的命题系统”（Gallagher，2022）。那么知识在定义上就是一张图，因此知识图谱只是一个同义反复。
According to Husserl, “knowledge […] can be understood as a system of interconnected propositions linked by inferential relations” (Gallagher, 2022). Then knowledge is a graph by definition, so knowledge graph is simply a tautology.
这是另一个观点。传递信息就是从内部改变状态。其原因——温度变化、肋骨刺痛、说话等——来自神经系统受到的刺激，但状态的变化是由生命系统的结构决定的，而不是由外部“信息”决定的。然而，结构耦合的历史[9]产生了协调，外部观察者可以将其视为从发送者到接收者的信息传递，但这样的事情根本不可能发生。
Here is yet another perspective. To inform is to change the state from within. The cause – a temperature change, a poke in the ribs, spoken words, etc. – is coming from irritation received by the nervous system, but the change of state is determined by the structure of the living system and not by the external “message.” Yet, the history of structural coupling [9] produces coordination that an external observer can perceive as a transfer of information from sender to receiver, but nothing like this can possibly happen.
只要语言被认为是外延性的，就必须将其视为一种信息传递的手段，就好像某种东西从一个有机体传递到另一个有机体，传递方式是“接收者”的不确定性范围应该根据“发送者”的规范而缩小。然而，当人们认识到语言是内涵性的而非外延性的，其功能是引导被引导者在其认知领域内进行引导，而不考虑引导者的认知领域时，显然语言并不能传递信息。被引导者有责任选择将他的认知领域引导到何处，这是他对自己状态的独立内部操作的结果；这种选择是由“信息”引起的，但由此产生的引导与“信息”对被引导者的意义无关。
So long as language is considered to be denotative it will be necessary to look at it as a means for the transmission of information, as if something were transmitted from organism to organism, in a manner such that the domain of uncertainties of the “receiver” should be reduced according to the specifications of the “sender”. However, when it is recognized that language is connotative and not denotative, and that its function is to orient the orientee within his cognitive domain without regard for the cognitive domain of the orienter, it becomes apparent that there is no transmission of information through language. It behooves the orientee, as a result of an independent internal operation upon his own state, to choose where to orient his cognitive domain; the choice is caused by the “message”, but the orientation thus produced is independent of what the “message” represents for the oriented.
（Maturana & Varela，1980）
(Maturana & Varela, 1980)
除了生物系统之外，这种对交流的理解在社会系统中具有各种含义，可以解释组织中决策的矛盾作用（Luhmann 等，2018）、组织之间的互动（Usher & Whitty，2017）以及科学与工业之间的生产性误解 [10]（Seidl，2010）。
Beyond biological systems, such understanding of communication has various implications in social systems for explaining, for example, the paradoxical role of decisions in organizations (Luhmann et al., 2018), interactions between organizations (Usher & Whitty, 2017), and the productive misunderstanding [10] between science and industry (Seidl, 2010).
如果生物系统和社会系统之间没有信息传递，那么知识也应该如此。那么，在引用的“知识图谱”定义中，“传达”一词的使用不能理解为机械传递。它应该从动态的角度来理解，即当用户与图谱交互时发生的事情。下一节将对此进行详细介绍。
If no transfer of information is going on between biological and social systems, then the same should apply to knowledge. Then the use of “convey” in the cited definition of “knowledge graphs” cannot be understood in the sense of a mechanical transfer. It should be understood in dynamical terms, as something happening when a user interacts with a graph. More on that in the next section.
根据 Maturana 和 Varela (1992) 的说法，
According again to Maturana and Varela (1992),
对是否存在知识的评估总是在关系背景下进行的。在这种背景下，扰动在生物体中引发的结构变化在观察者看来是对环境的一种影响。
the evaluation of whether or not there is knowledge is made always in a relational context. In that context, the structural changes which perturbations trigger in an organism appear to the observer as an effect upon the environment.
这并不是在定义什么是知识，而是在定义何时进行这样的评估。因此，它具有双重关系，首先与看似拥有知识的人的背景有关，然后与进行评估的观察者有关。然后，每种行为都可以被评估为一种认知行为，并且延伸而言，生活，因为它是“作为生命存在的有效行动”（Maturana & Varela，1992），等同于知识。
This doesn’t define what knowledge is, but when such evaluation is made. It is then relational in a double sense, once to the context of the one appearing to have knowledge and then to the observer making the assessment. Then every behavior can be assessed as a cognitive act and by extension, living, since it is “effective action in existence as a living being” (Maturana & Varela, 1992), is equivalent to knowing.
知识不能归因于非生命系统或物体。人可以知道，但图不能。[11] 尽管如此，谈论知识图谱还是有用的。虽然图谱不知道，但正如我们稍后会看到的那样，从它们所实现的扩展认知的角度来看，以这种方式引用知识图谱是有充分理由的。从这个角度来看，即使一般地谈论知识图谱有点问题，但对于 PKG 来说就没那么问题了。下一节将更详细地解释这一现象。现在，让我们继续探索知识图谱，回顾一下主要类型和一些突出的例子。
Knowledge cannot be attributed to nonliving systems or objects. People can know, but graphs cannot. [11] Still, speaking about knowledge graphs is useful. While graphs don’t know, as we’ll see later, there is a good reason to refer to knowledge graphs in that way from the perspective of the extended cognition that they enable. Seen through this lens, even if it is a bit problematic to talk about knowledge graphs in general, it is less so for PKGs. This phenomenon is explained in more detail in the next section. Now, let’s continue our exploration of knowledge graphs by reviewing the main types and some prominent examples.
知识图谱的类型
Types of knowledge graphs
我们可以区分三种类型的知识图谱：开放、企业和个人知识图谱。历史上首先出现的是开放知识图谱 (OKG)，例如 Freebase、DBpedia、Wikidata 和 Google Knowledge Graph。[12] OKG 可以进一步分为用于一般知识（或跨域）的 OKG，例如 Freebase、Wikidata、DBpedia 和 YAGO，以及用于特定领域的 OKG，例如 Uniprot 或 EU Publications (CELLAR)。
We can distinguish three types of knowledge graphs: open, enterprise, and personal knowledge graphs. Historically first came open knowledge graphs (OKG) such as Freebase, DBpedia, Wikidata, and the Google Knowledge Graph. [12] OKG can be further split into those for general knowledge (or cross-domain) like Freebase, Wikidata, DBpedia, and YAGO and domain-specific ones like Uniprot or EU Publications (CELLAR).
第二种类型是企业知识图谱 (EKG)。它们通过统一来自异构内部和外部数据源的数据，作为一种灵活的数据集成解决方案实现，或者用于服务于特定的业务案例，对于该业务案例，图形技术比其他替代方案提供更好的结果。企业知识图谱的两个主要驱动因素是来自传统的以应用程序为中心的软件系统设计和构建方法的问题，以及数据分析和机器学习日益增长的需求。近年来，来自不同行业的许多大公司都实施了企业知识图谱。突出的例子有宜家、Airbnb、罗氏、阿斯利康、拜耳、瑞银、西门子能源、富国银行、宝马、摩根士丹利和博世。企业知识图谱基金会 (EKGF) 最近成立。EKGF 发布了一套 EKG 原则和成熟度模型。[13]
The second type is the enterprise knowledge graphs (EKG). They are implemented as a flexible data-integration solution by unifying data from heterogeneous internal and external data sources or to serve a specific business case for which graph technologies give better results than their alternatives. The two main drivers for enterprise knowledge graphs are the problems coming from the traditional application-centric approach for designing and building software systems and the increased demands of data analytics and machine learning. Many big companies from different industries have implemented enterprise knowledge graphs in recent years. Prominent examples are IKEA, Airbnb, Roche, AstraZenekca, Bayer, UBS, Siemens Energy, Wells Fargo, BMW, Morgan Stanley, and Bosch. An Enterprise Knowledge Graph Foundation (EKGF) has been established recently. EKGF published a set of principles and a maturity model for EKG. [13]
个人知识图谱
Personal Knowledge Graphs
第三个也是最新的知识图谱是个人知识图谱 (PKG)。它们引起了学术界的关注，2019 年，Balog 和 Kenter 发表了一项研究议程，确定了未来工作的几个领域 (Balog & Kenter, 2019)。他们将 PKG 定义为
The third and most recent knowledge graph species are the personal knowledge graphs (PKG). They got some attention from academia and in 2019 Balog and Kenter published a research agenda identifying several areas of future work (Balog & Kenter, 2019). They defined PKG as
关于实体及其之间关系的结构化知识来源，其中实体及其之间的关系具有个人重要性，而非一般重要性。该图具有特定的“蜘蛛网”布局，其中图中的每个节点都连接到一个中心节点：用户。
a source of structured knowledge about entities and the relation between them, where the entities and the relations between them are of personal, rather than general, importance. The graph has a particular “spiderweb” layout, where every node in the graph is connected to one central node: the user.
（巴洛格和肯特，2019 年，第 218 页）
(Balog & Kenter, 2019, p. 218)
由于我们生活在大规模个性化的时代（Esposito，2022），因此知识图谱自然也用于此，因此本文阐明了个性化和个人知识图谱之间的区别。虽然后者本质上支持个性化，但它包含由用户创建或策划的“不相交的实体集”（Balog & Kenter，2019）。这些类型的图具有相同的首字母缩写词并共享一些功能，但性质不同。即使在科学出版物中，它们也很容易混淆。[14]
Since we live in times of mass personalization (Esposito, 2022) it is natural that knowledge graphs are also used for that, so the paper clarified the difference between personalized and personal KG. While the latter inherently enables personalization, it contains a “disjoint set of entities” (Balog & Kenter, 2019) created or curated by the user. These types of graphs have the same acronym and share some capabilities but are of different natures. They get easily confused even in scientific publications. [14]
这篇论文概述了四个研究问题。它们涉及知识表示、实体链接、图形填充和与外部资源集成等问题。这样说来，它们听起来并不是特别新颖，但它们是在内容由一个人创建并满足其需求时所面临的特定挑战的背景下提出的。
The paper outlined four research questions. They deal with problems such as knowledge representation, entity linking, graph population, and integration with external sources. Put this way they don’t sound especially new, but they are in the context of the specific challenges when the content is created by and for the needs of one person.
个人图谱填充问题似乎吸引了相对较多的研究工作，最近有关于从对话（Li et al., 2015; Mohanaraj & Laursen, nd; Tigunova et al., 2020; Torbati et al., 2021; Yu et al., 2020）、从非结构化文档（Vannur et al., 2020）和从异构数据源（De Mulder et al., 2021; Kalokyri et al., 2018; Montoya et al., 2018）中提取实体和关系的研究。De Mulder 等人的工作在两个方面具有特殊意义。与 Thymeflow（Montoya et al., 2018）类似，它生成具有明确语义的基于标准的 PKG，但此外，用户可以完全控制其数据，从而实现应用程序数据解耦和完全去中心化的架构。在讨论 PKG 的演变时，我们会回到这一点。
The personal graph population question seems to attract relatively more research efforts with recent contributions for entity and relationship extraction from conversations (Li et al., 2015; Mohanaraj & Laursen, n.d.; Tigunova et al., 2020; Torbati et al., 2021; Yu et al., 2020), from unstructured documents (Vannur et al., 2020) and from heterogeneous data sources (De Mulder et al., 2021; Kalokyri et al., 2018; Montoya et al., 2018). The work of De Mulder et al. has particular significance in two dimensions. Similar to Thymeflow (Montoya et al., 2018), it generates standards-based PKGs with explicit semantics, but in addition, the user has full control over their data, allowing application-data decoupling and fully decentralized architectures. We’ll get back to this point when discussing the evolution of PKGs.
推荐似乎是一个突出的研究课题。PKG 可以推荐饮食（Seneviratne 等人，2021 年）、活动（Safavi 等人，2020 年）或书籍、食谱和旅行目的地（Safavi 等人，2020 年）。
Recommendations seem to be a prominent research topic. The PKG can recommend a diet (Seneviratne et al., 2021), activities (Safavi et al., 2020) or books, recipes, and travel destinations (Safavi et al., 2020).
总体而言，事实证明，当前学术界的兴趣在于使用已经在图谱之外创建的数据，并且用户与图谱的交互强度相对较低，尤其是在写作和链接方面。因此，PKG 中的“个人”更多地用于表示有关个人的知识，而不是个人创建的知识。同时，在实践中，趋势恰恰相反。越来越多的人开始采用先进的网络笔记工具，即思维工具 (TfT)，它支持从日常计划和研究到项目管理和设计的广泛用例。他们中的许多人以某种方式将用户内容视为 PKG。我们将在后面关于 PKG 演变的部分中回顾如何以及在多大程度上进行这种做法。
Overall, it turns out that the current academic interest is in using data that is already created outside the graph, and there is a relatively low intensity of user interaction with the graph, especially in terms of writing and linking. “Personal” in PKG is then used more for knowledge about a person than created by a person. At the same time, in practice, the trends are just the opposite. There is growing adoption of advanced networked note-taking tools, referred to as Tools for Thought (TfT), supporting a wide range of use cases from daily planning and research to project management and design. One way or another many of them treat the user content as a PKG. We’ll review how and to what extent later in the section about the evolution of PKG.
许多新的 PKG 工具被用于支持研究。这并不奇怪。Luhmann 的 Zettelkasten 完全符合研究用 PKG 的条件（详情请参阅本卷中的其他章节），因此也是有记录以来使用时间最长的 PKG——约 45 年。其使用的好处在 Luhmann 工作的质量和数量以及 Luhmann 自己对其系统的贡献的评估以及他对它的依赖程度上显而易见（Luhmann，1981 年）。[15]
Many new PKG tools are used to support research. That’s not surprising. Luhmann’s Zettelkasten fully qualifies as a PKG (see my other chapter in this volume for details) for research and, as such is the longest used PKG on record – about 45 years. The benefits of its use are evident in the quality and quantity of Luhmann’s work and Luhmann’s own assessment of the contribution of his system and to what extent he relied on it (Luhmann, 1981). [15]
PKG，或者至少是那些用于个人知识管理的 PKG，在某种程度上始终具有社交性，因为用户可以与过去和未来的自己进行交流。一旦个人图谱的某些部分在不同用户之间共享，我们就可以称之为人际（Lajos，2019；Ivanec，本卷）知识图谱。一个密切相关的类别是协作知识图谱 (CKG)。一些当前的 PKG 工具具有协作功能。但是，可能存在基于知识图谱的协作平台，这些平台仅用于协作，在这种情况下，它们不符合 PKG 的资格。
PKGs, or at least those that are used for personal knowledge management, are in a way always already social since the user communicates with her past and futures selves. Once some parts of the personal graphs are shared among different users, we can speak of inter-personal (Lajos, 2019; Ivanec, this volume) knowledge graphs. A closely related category is collaborative knowledge graphs (CKG). Some current PKG tools have collaborative capabilities. However, there can be collaborative platforms based on knowledge graphs that are used only for collaboration, in which case they don’t qualify as PKGs.
在快速概述了 PKG 种类和用例之后，是时候回过头来提出一个定义。Balog 和 Kenter 之前提供的 PKG 定义来自 2019 年。
After this quick overview of the PKG species and use cases, it’s time to go back and propose a definition. The PKG definition provided earlier from Balog and Kenter is from 2019.
与此同时，Ruben Verbourgh 对 PKG 进行了非常全面的定义，即“您自己创建的所有数据与其他人创建的有关您的所有数据的结合。”（Global Data Geeks，2021 年）。
In the meantime, Ruben Verbourgh defined a PKG very inclusively as “All data that you yourself create combined with all data that others create about you.” (Global Data Geeks, 2021).
最近，Skjæveland 和 Balog 走向了另一个极端。他们修改了 2019 年的定义，并提出了更为严格的定义：
Recently, Skjæveland and Balog went to the opposite extreme. They revised the definition from 2019, and proposed a stricter one:
个人知识图谱 (PKG) 是一种知识图谱 (KG)，其中单个个人（称为 PKG 的所有者）拥有 (1) 对 KG 的完全读写访问权限，以及 (2) 授予其他人对 KG 任何指定部分的读写访问权限的专有权。PKG 的主要目的是支持交付专门为其所有者定制的服务。
A personal knowledge graph (PKG) is a knowledge graph (KG) where a single individual, called the owner of the PKG, has (1) full read and write access to the KG, and (2) the exclusive right to grant others read and write access to any specified part of the KG. The primary purpose of the PKG is to support the delivery of services that are customized particularly to its owner.
（Skjæveland 等人，2023 年，第 5 页）
(Skjæveland et al., 2023, p. 5)
虽然这个定义比 2019 年的定义有所改进，提供了明确的标准，并消除了事实需要与用户关联的不必要条件，但它跳过了“知识图谱”，完全关注数据所有权。数据所有权确实是我所说的“第三波”的一个重要方面和核心（请参阅下面关于 PKG 演变的部分）。然而，“个人”不应只局限于所有权这一方面。我建议通过对 Hogan 等人提出的知识图谱定义进行最小程度的修改来定义 PKG。（Hogan 等人，2022 年），基本上删除真实的 [16] 并添加个人。
While this definition is an improvement over the one from 2019 by providing clear criteria and removing the unnecessary condition that facts need to be connected to the user, it skips “knowledge graph”, and focuses entirely on data ownership. Data ownership is indeed an important aspect and central for what I call the “third wave” (see below the section about the evolution of PKG). Yet, “personal” shouldn’t be reduced to only one aspect, that of ownership. I would suggest defining PKGs by a minimal modification of the definition for knowledge graph proposed by Hogan et al. (Hogan et al., 2022), basically removing real [16] and adding personal.
用于积累和传达世界知识的数据图，其节点代表个人感兴趣的实体，其边代表这些实体之间的关系。
A graph of data intended to accumulate and convey knowledge of the world, whose nodes represent entities of personal interest and whose edges represent relations between these entities.
图形延伸思维
The graph-extended mind
自然认知系统……参与意义的产生……从事变革性的而非仅仅是信息的互动：它们塑造了一个世界。
Natural cognitive systems … participate in the generation of meaning … engaging in transformational and not merely informational interactions: they enact a world.
弗朗西斯科·瓦雷拉，《具身心智》，1991 年
Francisco Varela, The Embodied Mind, 1991
最简单的信息技术就是笔和纸。但是当我们使用它们时，事情并不简单。一旦我们在纸上做了一个标记 [17]，我们就开始交谈。这相当于口头交谈（Glanville，2007）。当我们看着标记看它暗示什么时，就像在口头交流中倾听一样（Glanville，2007）。这是一个持续的互动，一个反馈循环，看到我们所做的事情会影响我们接下来的想法和行动，进而调节接下来的想法和行动。重要的是，它可以帮助建立全新的联系，而如果没有积极参与我们周围的环境，这种联系是不可能的。
The simplest information technology is pen and paper. But when we use them what happens is not simple. Once we make a mark [17] on a piece of paper, we start a conversation. It is equivalent to talking in verbal conversation (Glanville, 2007). When we look at the mark to see what it suggests, it is like listening in verbal communication (Glanville, 2007). There is an ongoing interaction, a feedback loop where seeing what we have done is influencing what we think and do next and that in turn modulates the ideas and actions that follow. Importantly, it can help in making completely novel connections that wouldn’t be possible without active involvement with our surroundings.
我们生命体与环境之间的互动具有认知意义。沃森和克里克使用纸板模型的工作对 DNA 的发现至关重要（沃森，1968 年）。对于卢曼来说，“没有写作就不可能思考”（卢曼，1981 年，M. Kuehn 译）。与他的 Zettelkasten 的互动是他工作的关键。
The engagement between our living bodies and the environment has cognitive significance. Watson and Crick’s work with cardboard models was essential for the discovery of DNA (Watson, 1968). For Luhmann, it was “impossible to think without writing” (Luhmann, 1981, translated by M. Kuehn). The interaction with his Zettelkasten was key for his work.
那么，认知是发生在头骨内部的事情，还是延伸到环境中？认知是大脑内部过程的问题，还是从我们与世界的互动中产生的？
Then is cognition something that happens within the skull, or is it extended in the environment? Is it a matter of internal processes in the brain, or is it emerging from our interaction with the world?
越来越多的证据表明，认知不是操纵外部世界表征的问题，也不局限于头骨内，而是体现为（Varela 等，1991）、情境化（Robbins 和 Aydede，2008）、延伸（Clark，2003；Clark 和 Chalmers，1998）、分布化（Hutchins，1995、2010）的，并且生命、心智（Thompson，2007）和语言（Paolo 等，2018）之间存在连续性。
There is growing evidence that cognition is not a matter of manipulating representations of the external world and is not bounded within the skull but is embodied (Varela et al., 1991), situated (Robbins & Aydede, 2008), extended (Clark, 2003; Clark & Chalmers, 1998), distributed (Hutchins, 1995, 2010), and there is continuity between life, mind (Thompson, 2007), and language (Paolo et al., 2018).
稍微解释一下，广义上讲，认知科学的主要分歧在于计算范式和具身范式。第一个计算学派是认知主义，流行于 20 世纪 70 年代和 80 年代，但至今仍以不同的形式存在。它严肃地将计算机视为思维的隐喻，并将大脑视为独立于身体和环境的东西。认知主义者声称，大脑通过处理预先给定的世界的内部符号表征来工作。另一个计算学派联结主义使用神经网络来比喻大脑。它有一个更系统的观点，但仍然声称外部世界只能通过内部表征来了解，但类型不同。
To unpack this a bit, broadly speaking, the main divide in cognitive science is between the computational and the embodied paradigm. The first computational school of thought is Cognitivism, popular in the 1970s and ’80s but still present to this day in different forms. It takes seriously the computer as a metaphor for the mind and sees the brain as something independent from the body and the environment. The brain, cognitivists claim, works by processing internal symbolic representations of a pregiven world. The other computational school, Connectionism, uses the neural network as a metaphor for the brain. It has a more systemic view but still claims that the outside world is only known through internal representations but of a different kind.
简而言之，根据计算范式，认知：
In short, according to the computational paradigm, cognition:
发生在大脑中（神经中心主义）
Happens in the brain (neurocentrism)
运用外部世界的内部表征（表象主义）
Works with internal representations of the external world (representationalism)
脱离肉体，独立于环境而运作
Is disembodied and independent of the environment for its functioning
这三种理论承诺与 4E 认知的立场形成了鲜明的对比。[18] 该学派的一些框架，如扩展心智假设（Clark & Chalmers，1998 年）、分布式认知（Hutchins，1995 年）和实施（Di Paolo 等人，2017 年；Stewart 等人，2014 年；Thompson，2007 年；Varela 等人，1991 年）可以帮助理解用户与 PKG 的交互。
These three theoretical commitments are in stark contrast with the position of 4E cognition. [18] Some frameworks in this school, such as the extended mind hypothesis (Clark & Chalmers, 1998), distributed cognition (Hutchins, 1995) and enaction (Di Paolo et al., 2017; Stewart et al., 2014; Thompson, 2007; Varela et al., 1991) can help in understanding the interaction of a user with a PKG.
1998 年，安德里·克拉克 (Andry Clark) 和戴维·查尔默斯 (David Chalmers) 提出了扩展心智假说 (Clark & Chalmers, 1998)。根据该假说，心智并不存在于大脑和身体中，而是以依赖于行动的方式延伸到物理世界。作者将有机体和环境的互动视为一种双向互动，“创造了一个耦合系统，该系统本身可以看作是一个认知系统”(Clark & Chalmers, 1998)。[19] 克拉克和查尔默斯举了将形状装入插座、写作、玩拼字游戏和船舶导航等例子，参考了哈金斯关于分布式认知的研究 (Hutchins, 1995)。他们还建议对一个假设的人奥托进行一个思想实验，奥托患有阿尔茨海默病，他依赖笔记本的方式就像他依赖正常运作的生物记忆一样。这个 1998 年的例子涉及记忆和笔记本，从今天的角度来看，它具有额外的意义。在研究中，“记忆支持潜力被认为是 PKG 应用最有趣的方面之一”（Balog 等人，2022 年）。同时，在行业中，网络笔记的高级应用目前是 PKG 的主要用途。
The extended mind hypothesis was proposed by Andry Clark and David Chalmers in 1998 (Clark & Chalmers, 1998). According to that hypothesis, the mind does not reside in the brain and body but extends into the physical world in an action-dependent way. The authors view the engagement of organism and environment as a two-way interaction “creating a coupled system that can be seen as a cognitive system in its own right” (Clark & Chalmers, 1998). [19] Clark and Chalmers give examples with fitting shapes into sockets, writing, playing Scrabble, and ship navigation, referring to the work of Hutchins on distributed cognition (Hutchins, 1995). They also suggest a thought experiment with a hypothetical person Otto, suffering from Alzheimer’s disease, who relies on his notebook the way he would rely on his biological memory if it was well functioning. This example from 1998, involving memory and a notebook, gains additional significance from today’s perspective. In research “the potential for memory support, [is pointed at] as one of the most interesting aspects” (Balog et al., 2022) for application of a PKG. At the same time, in the industry, the advanced applications for networked note-taking are currently the dominant utilization of PKG.
对于 Hutchins 来说，问题不在于在哪里寻找认知，而是避免预先设定分析单位（Hutchins，2010）。对于某些现象，头骨可能是正确的边界，而对于其他现象则不是。他对分布式认知的研究表明，许多归因于个人的认知成就实际上来自超越个人大脑和身体的认知系统（Hutchins，1995），必须被视为“制定理解的系统”（Hutchins，2010）。这项工作有助于理解用户与 PKG 之间的交互，甚至更有助于理解 PKG 之间的交互。
For Hutchins, the question is not where to look for cognition but to avoid setting the unit of analysis in advance (Hutchins, 2010). For some phenomena, the skull can be the right boundary, for others not. His work on distributed cognition shows how many of the cognitive accomplishments attributed to individuals are in fact coming from cognitive systems that transcend individual brains and bodies (Hutchins, 1995) has to be seen as “system of enacted understandings” (Hutchins, 2010). This work is helpful for understanding the interaction between user and PKG and even more for inter-PKG.
这里特别令人感兴趣的是，当心理状态的出现是通过与万维网等复杂的技术网络进行互动时，向系统和社会技术视角的转变如何发挥作用 (Halpin 等人，2013 年；Smart，2013 年；Smart 等人，2010 年)。
Of particular interest here is how the shift to a systemic and socio-technical perspective plays out when what mediates the emergence of mental states is the interaction with complex technological networks like the World Wide Web (Halpin et al., 2013; Smart, 2013; Smart et al., 2010).
Smart 等人 (2010) 研究了当神经外资源是社会技术网络支持的环境时，扩展和分布式认知的含义。与此类环境的互动“不仅会增强或提高某些已建立的能力；它会产生全新形式的认知处理能力”（Smart 等人，2010）。他们提出了网络扩展思维的论点：
Smart et al. (2010) researched the implications of extended and distributed cognition when the extra-neural resources are socio-technical network-enabled environments. The interaction with such environments “does not merely result in the augmentation or enhancement of some well-established ability; it engenders entirely new forms of cognitive processing capability” (Smart et al., 2010). They proposed the thesis of the network-extended mind:
在某些情况下，大规模信息和通信网络的技术和信息元素可以构成主体心理状态和过程（至少是部分）的物质随附基础的一部分。
The technological and informational elements of large-scale information and communication networks can, under certain circumstances constitute part of the material supervenience base for (at least some of) an agent’s mental states and processes.
那么图扩展思维的概念将不再是本论文的隐含和扩展，而只是重新表述。事实上，Smart 等人明确指出了 RDF 和 OWL 等技术。
Then the notion of graph-extended mind will not be a matter of implication and extension of this thesis but mere rephrasing. In fact, Smart et al. explicitly point to technologies such as RDF and OWL.
遵循这一论点，可以将其解释为心智通过来自环境的物体（例如笔和纸、导航地图或 PKG）得到扩展。声称在某种程度上，这些物体本身构成了扩展。甚至可能有些作者已经接近这种理解。为了避免这种风险，最好将这种现象视为一种突发和实施的心智，而不是一种扩展的心智。
When following this argument, it is possible to interpret it as the mind being extended with objects from the environment, such as pen and paper, a navigation map, or a PKG. To claim that, in a way, these objects themselves constitute the extension. It might even be that some authors are coming close to that understanding. To avoid this risk, it is better to think of that phenomenon as an emergent and enacted mind rather than an extended one.
另一个风险是将其与“第二大脑”的概念联系起来，这一概念在蒂亚戈·福尔特（Forte，2022 年）的课程和著作之后变得流行起来。“打造第二大脑”是一个棘手的比喻，可能有助于提高生产力，但更认真地对待它需要对两种不相容的观点做出承诺，即神经中心主义和外在主义。
Another risk is to associate it with the idea of “second brain” that became popular after the courses and the book of Tiago Forte (Forte, 2022). “Building a second brain” is a sticky metaphor and might help in some practices for increasing productivity, but taking it more seriously would require a commitment to two incompatible views, neurocentrism and externalism.
更进一步说，我们有理由相信，这种通过与某些类型的工具的交互而出现的系统是自我维持和自主的，因此，它与生物系统和社会系统属于同一类系统。[20]
Taking a step further, there is a good reason to believe that such a system, emerging through interaction with certain types of tools, is self-sustained and autonomous, and as such, it’s in the same class of systems as biological and social systems. [20]
什么样的工具可以实现这一点？当然不是每种工具，也不是每种软件工具。未来研究的重点是确定是否有某种类型的能力可以增加这种认知系统出现的可能性。属于这一类的工具之一是电子游戏。
What kind of tools can make this happen? Certainly not every tool, and not every software tool. It is a matter of future research to confirm if there are types of capabilities that increase the likelihood of such a cognitive system to emerge. A type of tools that belongs to this class are videogames.
游戏玩法被认为是玩家与游戏之间二元和互惠耦合的成果。在这种互惠关系中，游戏玩法以自主组织的形式出现，既能自我维持，又不稳定。（Vahlo，2017 年）
[G]ameplay is argued as being the achievement of dyadic and reciprocal coupling between a player and the game. In this reciprocity, gameplay arises as autonomous organization that is both self-sustaining and precarious. (Vahlo, 2017)
另一种类型的工具很可能是 PKG。Luhmann 已经承认，在对他的 Zettelkasten 进行了大量工作之后，产生的东西“获得了自己的生命，独立于其作者”（Luhmann，1981）[21]。现在更先进的技术可以加速这一进程。
And most likely another type of such tools are PKGs. Already Luhmann admitted, that after some extensive work with his Zettelkasten, what arises “gets its own life, independent of its author” (Luhmann, 1981) [21]. The more advanced technologies available now can accelerate that process.
可以通过一组结构平衡来分析此类新兴系统的可行性，例如自主性和凝聚力之间的平衡（Velitchkov，2020 年）。这些系统并非存在于真空中，而是依赖于它们所嵌入的社会技术系统的可行性。但由于这些社会技术系统属于同一类，因此可以通过同一组平衡来分析它们的可行性。
The viability of such emergent systems can be analyzed through the lens of a small set of structural balances, like the one between autonomy and cohesion (Velitchkov, 2020). These systems don’t exist in a vacuum but are dependent on the viability of the socio-technical system they are embedded in. But since these socio-technical systems are of the same class, their viability can be analyzed by the same set of balances.
PKG 的演变
Evolution of PKGs
目前 PKG 的采用情况如何？它将如何变化？PKG 将如何发展？
What is the current adoption of PKGs? How will it change, and how will the PKG evolve?
诸如“谁目前正在使用 PKG？”之类的问题，根据解释，可以有两个极端的答案：
A question such as “Who is currently using a PKG?”, depending on the interpretation, can have two extreme answers:
还没有人使用 PKG。
Nobody is using a PKG yet.
每个人都已经在使用 PKG。
Everybody is already using a PKG.
如果我们对当前的状态应用某些资格标准，将得到接近第一个极端的答案。如果我们使用相同的标准，我们可以预期，未来会缓慢或迅速地走向第二个极端。但根据对 PKG 的解释方式，这两个答案都可以提供对当前状态的公平描述。更重要的是，这种解释显示了生态系统如何演变的有趣轨迹。
If we apply certain qualification criteria for the current state of play will give an answer close to the first extreme. If we use the same criteria, we can expect, slowly or quickly, to move to the second extreme in the future. But depending on how a PKG is interpreted, both answers can provide a fair description of the current state. More importantly, this interpretation shows an interesting trajectory of how the ecosystem is evolving.
第一波
The First Wave
当然，仅仅压缩是不够的；人们不仅需要制作和存储记录，还需要能够查阅它。
Mere compression, of course, is not enough; one needs not only to make and store a record but also be able to consult it.
万尼瓦尔·布什，《如我们所想》，1945 年
Vannevar Bush, As We May Think, 1945
自个人电脑诞生以来，某些需求（无论是真实的还是最初想象出来的）逐渐发展成为文本处理、电子表格和演示程序等应用类别。这些方便打包的解决方案使它们广受欢迎且需求旺盛，这反过来又进一步改善和巩固了它们。我们的生活越来越好，但其数字延伸却变得高度碎片化。
Since the dawn of personal computing, certain clusters of needs, real or first only imagined, crystalized into application classes such as text processing, spreadsheets, and presentation programs. Being conveniently packaged solutions made them popular and in demand, which in turn further improved and solidified them. Our life got better, but its digital extension grew highly fragmented.
如果我们看看我们处理的数据类型，就会发现每种数据类型都只由其专用的应用程序管理。我们将书签保存在浏览器或某些 Web 应用程序中以进行注释。我们使用文本处理应用程序创建文档，但在编写文档时，我们无法搜索书签。当我们与他人共享文档时，除非他们碰巧使用相同的文本编辑器，否则他们会遇到困难。[22] 我们的任务列表与书签和文档是分开的。一些应用程序提供邮件、任务和日历管理的组合，但如果我们有更复杂的计划方式，我们会选择专用的任务管理应用程序。传统的笔记应用程序通常具有类似的集成功能。它们可能提供 Web 剪辑功能，将笔记、书签和注释集成在一起。然而，我们有一个单独的应用程序来管理项目，另一个用于收集研究论文，一个用于电子表格和查看照片，一个用于数据库。我们的个人数据是断开连接的。每个应用程序和每个文档都是一个孤岛。但除了碎片化之外，个人信息管理还容易受到数字囤积和重新查找问题的困扰。
If we look at what types of data we work with, we see that each one is only managed by its dedicated application. We keep our bookmarks in a browser or in some web application for annotations. We create documents with our text processing application, but from there, while we write, we cannot search for our bookmarks. When we share a document with another person, they have difficulties with it unless they happen to use the same text editor. [22] Our task lists are separated from bookmarks and documents. Some applications offer a combination of mail, task, and calendar management, but if we have a more sophisticated way of planning our days, we opt for dedicated a task management application. Similar integration is often featured in traditional note-taking applications. They may offer a web-clipping capability bringing integration of notes, bookmarks, and annotations. Yet we have a separate application for managing projects, yet another for collecting research papers, one for spreadsheets and viewing photos, and one for databases. Our personal data is disconnected. Each application and each document is a silo. But apart from fragmentation, personal information management tends to suffer from digital hoarding and re-finding problems.
创建和存储照片、书签和笔记越容易，我们收集的信息就越多。渐渐地，我们变成了信息囤积者。我们经常会发现一些有价值的东西，可以暂时保留，但之后我们再也不会用到，要么是因为我们不需要它，要么是因为我们忘记了它，要么就是因为我们根本找不到它。因此，我们可以大致区分出四种情况：
The easier it was to create and store photos, bookmarks, and notes, the more we collected. Gradually we turned into information hoarders. It is often that we find something valuable to keep at the moment, and we never use it later either because we don’t need it, we forget about it or we simply cannot find it. Thus, we can broadly distinguish four cases:
某些东西我们在收集时会认为它有价值，但实际上它并无价值；
Something we believe is valuable at the moment we collect it, but it never actually is;
某些东西确实很有价值的，但我们在需要它时却很难找到它；
Something that’s actually valuable, but we have trouble finding it when we need it;
有些东西很珍贵，但我们却忘记了我们拥有它；
Something that’s valuable, but we forget we have it;
某些东西在特定环境下可能会很有价值，尤其是当它（半）意外地出现在那里时。
Something that might be valuable in a certain context, especially if it appears there (semi)unexpectedly.
其中一些情况在网上搜寻中很容易看到。我们想回到我们认为有价值的东西。而这样做的策略之一就是添加书签。这是重新查找网站的最快方法（O. Bergman & Whittaker，2016），但最近的一项研究表明，只有 16% 的研究样本采用这种方法，其余的人更喜欢替代策略（O. Bergman 等人，2021），例如定向越野（Teevan 等人，2004）[23] 或重现他们试图重新查找的信息的第一次体验（Jones 等人，2014）。
Some of these cases are easy to see in web foraging. We want to go back to something we have assessed as valuable. And one of the strategies to do so is bookmarking. It is the fastest method to refind websites (O. Bergman & Whittaker, 2016), and yet a recent study shows that it’s practiced only by 16% of the study sample, with the rest preferring alternative strategies (O. Bergman et al., 2021) like orienteering (Teevan et al., 2004) [23] or reenacting their first experience with the information they are trying to refind (Jones et al., 2014).
根据搜索内容，会采用不同的重新查找策略 [24]，[25] 但每种类型的数字对象都通过单独的工具或工具内单独的方式访问。我们使用操作系统的文件管理器查找文件。我们使用浏览器中的默认搜索引擎查找网络文档 [26]，然后使用不同的搜索方法查找书签。[27] 当我们在同一个文档中时，例如 Microsoft Word，我们无法从其他类型的文档（如演示文稿和电子表格）中查找信息。
Different strategies for refinding [24] are applied depending on what is searched, [25] but each type of digital object is accessed via a separate tool or a separate way within a tool. We look for files using the operating system’s file manager. We look for web documents using the default search engine in our browser [26] but then use a different search method to find bookmarks. [27] When we are in the same document, for example, Microsoft Word, we are not able to look for information from other types of documents like presentations and spreadsheets.
以下情况很常见。我们收到朋友关于特定主题的问题，我们碰巧知道有一本关于该主题的好书，但我们不记得这本书的名字。由于我们没有书名，搜索主题和与之相关的内容也无济于事。然后我们记得这是另一个朋友推荐的，但我们不记得在哪个渠道，所以我们去查看电子邮件、Messenger、Signal 和 Telegram 上与另一个朋友的消息。我们没有找到它。幸运的是，我们找到了另一条消息，提示我们查看手机上的短信。最后，它终于找到了。一个类似的常见情况是，当一个文件被共享时，我们找不到链接，但我们记得上下文——是在一次会议上——所以我们在日历中搜索，希望在那次会议邀请中找到它。
The following situation is common. We get a question from a friend about a specific topic, and we happen to know of a good book that addresses that topic, but we can’t remember the name of the book. Since we don’t have the book name, searching for the topic and something associated with it doesn’t help. Then we remember it was recommended by another friend, but we don’t remember on which channel, so we go and check the messages with that other friend on email, Messenger, Signal, and Telegram. We don’t find it. What we find instead, luckily, is another message that prompts us to check the text messages on the phone. And finally, there it is. A similar common case is when a file was shared, and we can’t find the link, but we remember the context – it was at a meeting – and so we search in our calendar in hope to find it in that meeting invitation.
所有这些不同类型的项目，电子邮件、联系人、日历事件、照片、网址、突出显示、段落、幻灯片、待办事项和项目问题，在关系网络中都是有价值的。而且关系的类型也不同。有些关系是项目之间的，比如电子邮件中的段落与幻灯片中的项目符号相关。其他关系是项目与创建它们或首次体验它们的环境之间的关系。可能最重要的是它们与新环境的关系，只要我们能轻松找到它们，它们就能带来价值。
All these different types of items, email, contact, calendar event, photo, web address, highlight, paragraph, slide, to-do item, and project issue, are valuable in the network of relations. And there are different types of relations. Some relations are between items, like a paragraph in an email is related to a bullet in a slide deck. Other relations are between the items and the context they were created in or otherwise first experienced in. Probably most important are their relations to new contexts, where they can bring value if only we can easily get to them.
我们确实会这样做，只要我们能将这些联系记在脑子里即可。
And we do, to the extent to which we can keep these links in our heads.
这就是为什么我们可以说每个人都已经在使用 PKG。
That’s why we can say that everybody uses a PKG already.
但这种 PKG 结构并不稳固。节点被困在应用程序孤岛和文件中。我们将边缘记在脑子里，或者通过重演之前的遭遇或采取小步骤缩小可能性来即时构建它们。
But that kind of PKG is a precarious structure. The nodes live trapped in application silos and files. We keep the edges in our heads or construct them on the fly by reenacting the previous encounter or by taking small steps to narrow down the possibilities.
从自主性-内聚性平衡的角度来看，我们的个人信息系统在应用程序内具有很高的内聚性。我们高度依赖（自主性降低）我们所使用的应用程序，因为我们喜欢它们，或者因为迁移成本高昂。同时，我们的数字环境并没有为我们的数据提供内聚性。相反，我们自己提供了一些内聚性，但仅限于我们的记忆和对手动操作的容忍度允许的范围内。
When seen through the lens of the autonomy–cohesion balance, our personal information system has a lot of cohesion within applications. We are highly dependent (reduced autonomy) on those we use because we like them or because the migration is costly. At the same time, our digital environment doesn’t provide cohesion for our data. Instead, we provide some cohesion ourselves but only to the extent allowed by our memory and by our tolerance for manual actions.
我建议将第一波 PKG 称为准 PKG。节点已存储，但它们没有全局标识符，只能从专用应用程序使用。另一个应用程序无法为同一类型的资源提供相同的功能。边缘不是以电子方式存储的。我们保留它们或按照某种关联链构建它们，无论是否有来自我们物理或数字环境的各种提示的帮助。
I’d suggest calling this first wave of PKG, quazi-PKG. The nodes are stored, but they don’t have global identifiers and can only be used from a dedicated application. Another application is not able to provide the same functionality even to a resource of the same type. The edges are not stored electronically. We keep them or construct them following some chain of associations with or without the help of various cues from our physical or digital environment.
准 PKG 浪潮表明了个人计算的不足。毫不奇怪，后续浪潮中的一些努力正是应用 PKG 来解决这一不足（Montoya 等人，2018 年；Obenauer，2021 年；Safavi 等人，2020 年；Rosenauer，本卷）。
The wave of the quazi-PKG shows a deficiency in personal computing. It is not surprising that some of the efforts of the following waves apply PKG to address exactly this deficiency (Montoya et al., 2018; Obenauer, 2021; Safavi et al., 2020; Rosenauer, this volume).
第二次浪潮
The Second Wave
每个注释的质量仅来自系统内的链接和反向链接网络。
Every note receives its quality only from the network of links and back-links within the system.
尼克拉斯·卢曼，《通过滑盒交流》，1981 年
Niklas Luhmann, Communicating with Slip Boxes, 1981
世界变得越来越复杂和不可预测。刺激的数量和种类都越来越多，需要平衡它们并做出相应的反应 (Velitchkov, 2020)。成功应对世界取决于拥有必要的多样性 (Ashby, 1958/1991, 1956/2015)。这可以合理地解释为什么在 2020-21 年期间，网络笔记工具（通常称为“思维工具”）大量涌现 [28]。就像新的病毒变种不断出现和传播一样，几乎每个月都会出现新的 TfT 变种，其中一些吸引了大量用户并在短时间内创建了整个生态系统。除了没有额外参与的用户外，生态系统还包括博主、时事通讯作者、制作带有技巧和窍门的视频的人、培训师、贡献者以及扩展和主题的开发人员。许多 TfT 都以某种方式将数据视为知识图谱。
The world gets more complex and unpredictable. There are more stimuli, both in number and variety, and they need to be balanced with matching responses (Velitchkov, 2020). Dealing successfully with the world depends on having requisite variety (Ashby, 1958/1991, 1956/2015). That can be one plausible explanation for why in the period 2020–21 there was an eruption [28] of tools for networked note-taking, often referred to as “Tools for Thought.” Just like the new virus variants kept emerging and spreading, new variants of TfT popped up almost every month, some attracting lots of users and creating whole ecosystems in no time. Besides users with no additional involvement, the ecosystems include bloggers, newsletter writers, people making videos with tips and tricks, trainers, contributors, and developers of extensions and themes. Many of these TfTs treat the data, in one way or another, as a knowledge graph.
这个新市场的诞生与学术界对 PKG 的兴趣日益增加同时发生，但两者之间几乎没有任何影响。
The birth of this new market went in parallel with increased interest in PKGs by academia, with little if any influence between the two.
这是第二波，我建议称之为原始 PKG。[29]
This is the second wave, which I suggest calling proto-PKG. [29]
这些工具是否符合 PKG 的条件？我们无法给出简单的“是/否”答案。我们不应该使用布尔逻辑，而应该使用模糊逻辑，其中某些工具比其他工具更属于该集合。
Do these tools qualify as PKGs? We can’t give a simple yes/no answer. Instead of Boolean, we should apply fuzzy logic here, where some tools are more members of the set than others.
LinkedDataHub（Jusevičius，本卷）、ImageSnippets（Warren，本卷）等工具位于中心，还有 Thymeflow（Montoya 等人，2018）和本章前面提到的其他研究原型。Codex 的位置也位于中心，它是“一种通过隔离属性注释集成文本即图元模型的文本即图解决方案”（Palladino 等人，2020 年）使用 Neo4j，ixnote 使用用 Swift 编写的专有图形数据库。非常接近中心的是 Kanopi。它不使用 RDF 或 LPG 技术，而是在 PostgreSQL 之上概念上实现了 RDF 模型。它可以从网站中提取 RDF，然后将其转换为内部数据模型。数据库设计不会通过为每种实体类型使用一个表来创建模式依赖关系。相反，只有[30]两个表，一个用于节点，另一个用于边。
Tools such as LinkedDataHub (Jusevičius, this volume), ImageSnippets (Warren, this volume) are in the center, along with Thymeflow (Montoya et al., 2018) and the other research prototypes mentioned earlier in this chapter. The place of Codex, “a text-as-graph solution that integrates a text-as-graph meta model via standoff property annotations” (Palladino et al., 2020) using Neo4j, and ixnote using a proprietary graph DB written in Swift, is in the center too. Very close to the center is Kanopi. It doesn’t use RDF or LPG technology, but conceptually implements the RDF model on top of PostgreSQL. It can pull RDF from websites, and then translate it into the internal data model. The database design does not create schema dependency by using a table for each entity type. Instead, there are only [30] two tables, one for nodes and another for edges.
图1.1 Kanopi的数据模型（简化）
Figure 1.1 Data model of Kanopi (simplified)
稍远一点是 TfT 家族，它们共享类似的方法、架构、技术和功能，但后端和业务模型不同。该家族包括 Roam Research、Logseq、Athens Research [31] 和 Hulunote。它们都是用 Clojure 编写的，使用 Datalog 进行查询，使用 Datascript 作为客户端数据库。
A bit further from the center is a family of TfTs, sharing a similar approach, architecture, technologies and capabilities but having different backend and business model. The family includes Roam Research, Logseq, Athens Research, [31] and Hulunote. They are all written in Clojure, use Datalog for querying and Datascript as a client-side database.
由于该家族的所有成员都遵循了 Roam Research 的开创性工作，因此快速浏览一下其数据模型就足以说明为什么它们都符合 PKG 工具的资格。
Since all in that family followed the pioneering work of Roam Research, having a quick look at its data model will be sufficient to show why all of them qualify as PKG tools.
图中有两种主要类型的节点可以与 :block/children 关系相关联。[32] 一种称为“页面”，在 UI 中用作“块”的容器，这是第二种类型的节点。块节点可以与其他节点具有相同类型的关系，在 UI 中显示为嵌套块的层次结构。块的实际内容是一个字符串，可以看作是关系 :block/string 中的另一种节点类型。这个字符串，即块的内容，可以包含对页面和节点的引用。
There are two main types of nodes in the graph that can be related with :block/children relationship. [32] One is called “page” and in the UI serves as a container of “blocks,” which is the second type of node. Block nodes can have the same type of relationship with other nodes, appearing in the UI as a hierarchy of nested blocks. The actual content of a block is a string, and can be seen as another node type, along the relationship :block/string. This string, the content of a block, can contain references to both pages and nodes.
图1.2 Roam Research数据模型（简化）
Figure 1.2 Data model of Roam Research (simplified)
将这组工具视为 PKG 集中心的另一个原因是它们使用 Datalog。Datalog 是一种声明性编程语言，其图形查询功能与 SPARQL 类似。基本查询模式甚至在语法上也相似：[?s :block/children ?o] (Datalog) 和 {?s :hasChild ?o}(SPARQL)。
Another reason to consider this set of tools close to the center of the PKG set is that they use Datalog. Datalog is a declarative programing language and its graph querying capabilities are similar to SPARQL. The basic query patterns are even syntactically similar: [?s :block/children ?o] (Datalog) and {?s :hasChild ?o}(SPARQL).
从“页面”的角度来看，Roam Research 和 Logseq 等工具的数据结构可以概括为超图，其中块充当超边的角色（见图 1.3）。
The data structure of tools like Roam Research and Logseq can be generalized as a hyper-graph, from the perspective of “pages”, where blocks play the role of hyper-edges (see figure 1.3).
图 1.3 以块为中心的 TfT，例如 Roam Research 和 Logseq，可以概括为超图。
Figure 1.3 Block-centric TfTs like Roam Research and Logseq, can be generalized as hyper-graphs.
Roam 系列中的工具显然是 PKG，但离集合的中心稍远。虽然用户在处理非结构化和半结构化数据时具有很大的灵活性，但他们无法指定节点和边的类型，并且总体上无法提供更明确的机器可处理语义。它只能通过结合约定和规则 [33] 或通过扩展来部分实现。[34] 集合中心的 PKG 工具可以通过 RDFS/OWL 本体（De Mulder 等人，2021 年；Seneviratne 等人，2021 年）、“面向方面”的本体（Palladino 等人，2020 年）或其他方式提供显式语义。第二组（和第三组）中的 PKG 工具目前不允许使用外部本体，也不提供用户创建和使用自己的本体的方法。它通常被认为是不必要的，并且是一种额外的负担。然而，Tana 和 Capacities 等工具却证明事实并非如此。
The tools in the Roam family are clearly PKGs, yet a little away from the center of the set. While users have a lot of flexibility when working with nonstructured and semistructured data, they are not able, for example, to specify types of nodes and edges and overall don’t have a way to provide more explicit machine-processable semantics. It can be only partially achieved through combinations of conventions and rules [33] or through extensions. [34] The PKG tools in the center of the set can provide explicit semantics via RDFS/OWL ontologies (De Mulder et al., 2021; Seneviratne et al., 2021), “aspect-oriented” ontology (Palladino et al., 2020), or in other ways. The PKG tools from the second group (and from the third) don’t allow the use of external ontologies at the moment and don’t provide ways for the user to create and use their own. It is often deemed unnecessary and as an additional overload. Yet tools like Tana and Capacities proved otherwise.
转向协作和人际知识图谱（该组中的一些工具已经提供了此类功能），缺乏本体或其他提供共享意义的方法限制了协调和查询的潜力。此外，它还限制了在协作空间之外创建的内容的有效重用，无论是在用户的 PKG 中还是在 OKG 中。
Moving to collaborative and interpersonal knowledge graphs – and some tools in this group provide such capabilities already – the lack of ontologies or other means for providing shared meaning limits the potential of coordination and querying. In addition, it limits the effective reuse of content created outside the collaborative space, either in PKGs of the users or from OKGs.
距离 PKG 类中心较远的是基于 Markdown 的工具，例如 Obsidian、Foam、Dendron、Zettlr、nb、Emanote、Cosma、Bangle.io 和 Nota。他们的共同点是使用本地 Markdown 文件。[35] 这些工具可以被看作具有概念图模型，其中每个 Markdown 文件都是连接到其他文件的节点。用户在文件文本中使用 wikilinks 的常用语法创建连接：双方括号。每种工具都提供了一种保存节点元数据的方法，通常编码为前置 YAML，这是 Jekyll 首创的方法。其中一些工具（例如 Obsidian）允许用户以与第二组类似的方式识别和引用文本中的块。然而，所有这些应用程序都远离 PKG 集的中心，主要有两个原因。它们没有像以前的群体那样拥有那么多的“结构化知识”（参见第二部分中引用的 Balog 和 Kenter 的定义）和图形粒度（图形处于页面级别，而不是段落级别），并且它们不将关系视为一等公民。
Further from the center of the PKG class are Markdown-based tools like Obsidian, Foam, Dendron, Zettlr, nb, Emanote, Cosma, Bangle.io and Nota. What they have in common is their use of local Markdown files. [35] These tools can be seen as having a conceptual graph model where each Markdown file is a node connected to other files. The connections are created by the user in the text of the file using the common syntax of wikilinks: double square brackets. Each tool provides a way to keep node metadata, often encoded as front-matter YAML, an approach pioneered by Jekyll. Some of these tools, such as Obsidian, allow the user to identify and refer to blocks within the text in a similar way as the second group. Yet, all these applications are further away from the center of the PKG set for two main reasons. They don’t have as much “structured knowledge” (see the definition of Balog and Kenter, quoted in the second section) as the previous groups and graph granularity (the graph is at the level of the pages, not paragraphs), and they don’t treat relationships as a first-class citizen.
除了这三类工具外，还有其他工具可以被视为 PKG，它们中的大多数都远离该集合的中心。它们包括 Cmap [36] 和 The Brain 等老牌工具，以及 Heptabase 和 Scrintal 等新晋工具，其中的可视化图形不是文本输入视图的问题，而是用户创建内容的方式。
Apart from these three groups, there are other tools that can be regarded as PKGs, most of them distant from the center of the set. They include old ones such as Cmap [36] and The Brain and newcomers like Heptabase and Scrintal, where the visual graph is not a matter of a view over the textual input but is how the user creates the content.
第二波浪潮的主要成就和潜在好处是什么？
What are the main achievements and potential benefits of the second wave?
与第一波不同，想法与其他个人数据项之间的许多关系都是明确的。内聚力得到了增强，但并没有以过多的限制为代价：有灵活性，使用户能够以不同的方式利用和探索[37]他们的图表，以适应他们的工作风格。这些 PKG 工具支持学习、日记、研究、非线性写作、项目管理、个人关系管理、时间规划和记录、创意、协作、推荐、会议管理等用例。在不同上下文中重用同一节点[38]并在不同时间重新显示它的能力为用户与其图表之间的交互提供了额外的好处。
Unlike the first wave, many of the relationships between ideas and other personal data items are explicit. The cohesion is increased but not at the price of too many restrictions: there are affordances providing flexibility so that the user can both exploit and explore [37] their graphs in different ways, adapted to their work styles. These PKG tools support use cases such as learning, journaling, research, nonlinear writing, project management, personal relationship management, time planning and logging, ideation, collaboration, recommendations, meeting management, and many more. The ability to reuse the same node in a different context [38] and to resurface it at a different time provides additional benefits from the interaction between the user and their graph.
这一代 PKG 已经带来了许多独特的好处，并且有可能带来更多好处。使用 PKG 可以培养新的工作和思维习惯。现在我们甚至谈论“图形思维”（Gosnell & Broecheler，2020 年）。图形思维可能有助于处理复杂问题（Nathan，2021 年），它可以促进 EKG 和链接数据的采用（Velitchkov，2021a），并导致对某些非图形工具的功能进行扩展 [39]，以便将它们用作图形工具 [40]，并创建提供类似图形体验的扩展。[41] 此类实践的激增可能会导致这些工具向知识图谱发展。
This generation of PKGs has brought many unique benefits already and has the potential to bring more. Working with PKGs cultivates new working and thinking habits. Now we even talk about “graph-thinking” (Gosnell & Broecheler, 2020). Thinking in graphs may help in dealing with complex problems (Nathan, 2021), it can contribute to the adoption of EKG and Linked Data (Velitchkov, 2021a) and it leads to exapting [39] of features of some non-graph tools so that they are used as if they are graph tools [40] and creating extensions providing graph-like experience. [41] A proliferation of such practices may lead to the evolution of these tools towards knowledge graphs.
第二波 PKG 的局限性是什么？
What are the limitations of the PKGs in the second wave?
这些 PKG 工具更有利于创建互联的个人数据并集成更多在工具之外创建的数据。然而，即使有了它们，我们的大多数个人数据仍然在 PKG 之外，驻留在电子邮件、电子表格和幻灯片中，或者在社交媒体平台、生活记录和医疗保健数据存储中。例如，如果我想在我的个人数据中搜索“PKG”，我希望从我的文本文档、PDF、电子邮件、电子书库、LinkedIn、Twitter、Telegram、Signal、Discord 等对话中获得相关结果。使用当前的工具，PKG 与应用程序耦合，外部工具使用自己的存储，我无法做到这一点。
These PKG tools are more inviting to create interconnected personal data and to integrate more of the ones created outside the tool. Yet, even with them, most of our personal data is still outside the PKG, residing in emails, spreadsheets, and slide decks or in social media platforms, lifelogging and healthcare data stores. For example, if I want to search for “PKG” in my personal data, I would like to get relevant results from my text documents, PDFs, email, eBooks library, conversations in LinkedIn, Twitter, Telegram, Signal, Discord, and so on. With the current tools, where the PKG is coupled with the application, and the external ones are using their own storage, I can’t.
从这个意义上来说，“还没有人使用 PKG。”
In that sense, “Nobody is using PKG yet.”
第二波的所有这些局限性的根源在于，当前的 PKG 工具主要是以应用程序为中心的。它们既不会取代其他个人数据应用程序的功能或协作平台的服务，也不会取代那些使用用户控制的存储的应用程序和平台。我们需要的是更灵活、以人为本的实现互操作性（凝聚力）的方式，同时允许自由（自主）选择和组合管理个人数据的应用程序和服务。
The origin of all these limitations of the second wave is that the current PKG tools are predominantly application centric. They neither replace the functionality of other applications for personal data or the services of the collaborative platforms, nor those applications and platforms that use storage controlled by the users. What is needed is more flexible, person-centric ways of achieving interoperability (cohesion), while allowing freedom (autonomy) for choosing and combining applications and services managing personal data.
第三次浪潮
The Third Wave
数据是自我描述的，不依赖于应用程序的解释和含义。
Data is self-describing and does not rely on an application for interpretation and meaning.
以数据为中心的宣言
Data-centric Manifesto
第三个的好名字可以是“PKG”，但如果我们必须选择一个前缀，就像其他两个一样，它将是“d-PKG”。字母“d”可以表示分散、解耦和以数据为中心中的任何一种，或全部。
A good name for the third would simply be “PKG” but if we have to pick a prefix, like it was for the other two, it will be “d-PKG.” That letter “d” can suggest any of decentralized, de-coupled, and data-centric, or all of them.
对于第三次浪潮，迹象和因素不仅来自 PKG 的演变，还来自个人知识管理日益脱钩的历史趋势（参见第二章“脱钩”一节），更普遍的是来自对网络和企业 IT 现状日益增长的不满。如果第三次浪潮将按照这里设想的方式发生，那将是所有这些因素相互作用的结果，因此值得勾勒出更大的背景。
For the third wave, there are signs and factors coming not just from the evolution of PKGs, but also from the historical trend for increased decoupling in personal knowledge management (see the section “Decoupling” in the second chapter), and more generally from the growing discontent with the current state of the web and enterprise IT. If the third wave is going to happen the way it is envisaged here, it will be due to the interplay of all these factors, so it is worth sketching the larger context.
网络被设计成一个去中心化的系统，通过对一些标准（主要是 HTTP 和 HTML）的一致认可，人们可以自由选择几乎所有其他东西。人们终于可以自由表达自己，选择从哪里和如何获取信息。他们可以自由创新，构建新的浏览器、网站以及他们能想到的任何网络应用程序和服务。像这样的系统，如果拥有一个自我维护的组织，就可以很好地运行，并且具有良性循环的自然趋势。换句话说，它可以放大善，并开发自己的免疫系统来抵御任何威胁其生存能力的事物。它所需要的只是拥有正确的支持约束，例如上述标准，并允许所有子系统自主。这又是自主性和凝聚力之间的平衡（Velitchkov，2020 年）。它适用于动物、人类、部落、组织、社会以及网络等社会技术系统。
The web was designed to be a decentralized system where the agreement on a few standards, basically HTTP and HTML, enabled free choice on just about anything else. People were finally free to express themselves and to choose from where and how to get information. They were free to innovate on building new browsers, websites, and whatever web applications and services they could think of. A system like this, with a self-maintained organization, can work well and has a natural tendency for virtuous cycles. In other words, it can amplify goodness and develop its own immune system for whatever threatens its viability. All it needs is to have the right kind of enabling constraints, for example, the standards mentioned above, and to allow autonomy of all subsystems. This is again the balance between autonomy and cohesion (Velitchkov, 2020). It works for animals, people, tribes, organizations, society, and for socio-technical systems like the web.
网络作为一个去中心化系统蓬勃发展，人们可以自由选择并创造更多选择。然后，有一天，平台出现了。它们免费提供优质服务。或者至少一开始它们看起来不错，而且是免费的。实际上，它们既不好也不免费。这些平台远不如之前的去中心化网络那么好的信息提供者。我们看到的不是我们想要的，而是他们的算法决定[42]向我们展示的。这些平台的服务并不免费。恰恰相反。事实上，我们付了两次钱。第一次是成为他们的内容提供者，第二次是向他们提供我们的个人数据。重要的是，我们不仅向他们提供我们当前的个人数据，还向他们提供未来的数据，允许他们跟踪我们的在线行为。就这样，网络，一个由用户塑造的去中心化系统，变成了一个由少数强大公司塑造的超中心化[43]系统。它也形成了用户的期望。2019 年，Facebook 和谷歌宣布现在可以将图片从 Facebook 复制到 Google Photos。这是创新的新常态。正如鲁本·维博格 (Ruben Verborgh) 指出的那样，在能够将视频信号发送至 380,000 公里的距离 50 年后，我们庆祝终于能够将照片移动 11 公里 [44] (Verborgh, 2020)。
And the web flourished as a decentralized system where people were free to choose and create more choices. And then, one day, the platforms appeared. They offered good services for free. Or at least they looked good and free at first. In reality, they were (and are) neither good nor free. The platforms are not nearly as good information providers as the decentralized web before them. What we see is not what we are looking for but what their algorithms decide [42] to show us. And the services of these platforms are not free. Quite the contrary. In fact, we pay twice. Once by being their content providers and a second time by giving them our personal data. Importantly, we don’t give them only our current personal data but also future data by allowing them to track our online behavior. In this way, the web, a decentralized system, shaped by the users, turned into a hyper-centralized [43] system, shaped by a few powerful corporations. It also formed users’ expectations. In 2019 Facebook and Google announced that it was now possible to copy images from Facebook to Google Photos. That’s the new norm for innovation. As Ruben Verborgh pointed out, 50 years after being able to send video signals over a distance of 380,000km, we celebrate that we can finally move a photo by 11km [44] (Verborgh, 2020).
在某些情况下，中心化可能对互联网有益，[45] 但在内容和服务层面，它会造成全球性损害，包括限制多元化和竞争、扼杀创新（Thierer，2016 年；Verborgh，2020 年）、传播虚假新闻和操纵社会（Cadwalladr & Graham-Harrison，2018 年；Christl，2017 年；Zuboff，2019 年；McNamee，2019 年）。
Centralization may be beneficial for the internet in a few cases, [45] but when it comes to the layer of content and services, it creates global damage ranging from limiting pluralism and competition to suffocating innovation (Thierer, 2016; Verborgh, 2020), spreading fake news, and manipulation of society (Cadwalladr & Graham-Harrison, 2018; Christl, 2017; Zuboff, 2019; McNamee, 2019).
几十年来，企业应用程序的构建方式是，每个应用程序的数据模型都是独立的，被困在应用程序内部，而数据的解释则在应用程序代码中。这导致了高昂的变更成本和高昂的数据集成成本（McComb，2018 年）。这种传统的、以应用程序为中心的应用程序构建方式至今仍占主导地位。大多数大型企业都有数千个应用程序孤岛。他们试图通过数据仓库、数据湖、点对点接口和 API 来集成数据。所有这些方法都提供了部分和临时的解决方案，并增加了技术负担。
In enterprises, for decades applications were built in a way where the data model is separate for each application, trapped inside it, and the interpretation of the data is in the application code. This led to a high cost of change and a high cost of data integration (McComb, 2018). Such a traditional, application-centric way of building applications is dominant to this day. Most big enterprises have thousands of application silos. They try to integrate the data through data warehouses, data lakes, point-to-point interfaces, and APIs. All these methods provide partial and temporary solutions and add to the technical debt.
这些问题的共同点在于，无论是在网络还是在企业中，数据都与应用程序和平台紧密耦合。因此，走向解耦的趋势最能体现在以数据为中心的原则上（《以数据为中心的宣言》，第 2017 年）：
What is common in these problems, on the web and in enterprises, is that data is tightly coupled with applications and platforms. Hence the movements toward decoupling, best captured by the data-centric principles (Data-Centric Manifesto, n.d.):
数据是任何个人、组织和社会的重要资产。
Data is a key asset of any person, organization, and society.
数据是自我描述的，不依赖于应用程序的解释和含义。
Data is self-describing and does not rely on an application for interpretation and meaning.
数据以开放、非专有的格式表达。
Data is expressed in open, non-proprietary formats.
数据的访问和安全是企业数据层或个人数据库的责任，而不是由应用程序管理。
Access to and security of the data is a responsibility of the enterprise data layer or the personal data vault, and not managed by applications.
应用程序可以访问数据，发挥其魔力，并将其处理的结果表达回数据层。
Applications are allowed to visit the data, perform their magic and express the results of their process back into the data layer.
网络去中心化运动带来了自己的一套原则，即“提倡共存和互操作性，不鼓励封闭式网络”。[46] 争取互操作性符合以数据为中心的宣言的第三项原则和 FAIR 原则。[47]
The movements for web decentralization bring their own set of principles, to “urge coexistence and interoperability, and discourage walled gardens”. [46] Striving for interoperability is in line with the third principle of the Data-Centric Manifesto and with the FAIR principles. [47]
数据与应用程序、平台或主机之间似乎都存在着相同的紧密耦合模式，这种模式体现在各个层面。网络中的“围墙花园”（Berners-Lee，2009）就像企业中的应用程序孤岛，也像第一波中的个人应用程序和文件，以及第二波中的大多数 PKG 工具。
It seems the same pattern of tight coupling between data and application, platform or host, is manifested in all levels. The “walled gardens” (Berners-Lee, 2009) in the web are like the application silos in enterprises and like the personal applications and files from the first wave, as well as most of the PKG tools from the second wave.
图 1.4 从孤立的应用和文件，通过强大的公钥管理系统，到完全由所有者控制的解耦公钥管理系统
Figure 1.4 From silo apps and files, through powerful PKM systems, to decoupled PKG, fully controlled by their owners
第三次浪潮的一个主题是解耦。[48] 重点是应用程序与数据的解耦，以及主机与内容之间的解耦（Jacobson 等，2012）以及功能的解耦：
A somewhat uniting theme for the third wave then is decoupling. [48] The focus is decoupling application and data, but also decoupling between host and content (Jacobson et al., 2012) and decoupling of functionalities:
我不需要选择单个电子邮件客户端，而是可以从收件箱、撰写窗口和垃圾邮件过滤器中编写我最喜欢的电子邮件客户端吗？
Instead of needing to pick a single email client, can I compose my favorite email client out of an inbox, a compose window, and a spam filter?
（Litt，2021 年）
(Litt, 2021)
PKG 在第三次浪潮中会是什么样子？在这里，我们终于可以单独讨论 PKG 本身，而不是以某种方式使用知识图谱的个人知识管理工具（第二次浪潮）。它可能不是单一的，而是分布式的存储，但具有一致性，带来与在单个存储中相同的体验。无论如何，数据完全由用户拥有和控制，允许应用程序组件以用户控制的方式使用和操作它。这还将实现独立于平台、应用程序甚至主机的 PKG 间。
What will PKG look like in the third wave? Here we can finally talk about PKGs by themselves and not about tools for personal knowledge management that use knowledge graphs in some way (the second wave). It might not be a single but distributed storage, yet with a coherence bringing the same experience as if it’s in a single store. In any case, the data is fully owned and controlled by the user, allowing application components to use and manipulate it in a way controlled by the user. This will also enable platform-, application-, and possibly also host-independent inter-PKGs.
目前有哪些迹象表明第三波疫情正在开始？
What are the current signs the third wave is starting?
一个标志是 Solid 项目，它是第三次浪潮中 PKG 的一个典型例子。在 Solid 架构中，用户 PKG 驻留在称为“pod”的商店提供商中。用户可以自由选择 pod 提供商，并可以授予对不同应用程序的访问权限。仍有许多挑战，但该项目似乎已经超越了研究和原型设计阶段，并且有一些生产性使用报告（社交电视和数据的未来，2022 年）。佛兰德斯政府最近决定向公民提供 Solid pod（佛兰德斯政府和 Solid，2020 年），如果成功实施，可能会进一步加速所需技术的开发和更广泛的采用。
One sign is the Solid project, which is a prime example of PKGs in the third wave. In a Solid architecture, the user PKG resides in store providers called “pods”. Users freely choose the pod providers and can grant access to different applications. There are still many challenges, but it seems that the project is going beyond the research and prototyping phase and there are some reports of productive use (Social TV and the Future of Data, 2022). The recent decision of the Flemish government to provide Solid pods to citizens (The Flanders Government and Solid, 2020), if successfully implemented, might further accelerate the development of the needed technologies and wider adoption.
第二个迹象来自当前本地优先的 PKM 工具趋势。其中大多数是基于 Markdown 的网络笔记工具。尽管可以说 Markdown 不是真正的标准，存在多种风格并且存在多个问题（Melvær，2022），但它的广泛采用超出了开发者社区。第二波第三组的所有工具都使用 Markdown，遵循本地优先原则。这样，用户的内容（除了一些语法差异，例如块 ID 和文档元数据）与应用程序无关。它可以从一个应用程序（例如 Obsidian）查看和操作，然后用户可以继续使用 Logseq 或 Foam 进行编辑以使用特定功能。这已经是数据应用程序解耦和用户完全控制其数据的一个例子。Logseq 更进一步，提供了一种独立于 Logseq 应用程序查询数据的方法，并将图形转换为 RDF。[49]
A second sign is coming from the current trend for local-first PKM tools. The majority of those are Markdown-based tools for networked note-taking. Although it can be argued that Markdown is not a real standard, exists in many flavors and has multiple issues (Melvær, 2022), it has wide adoption that goes beyond the developer community. All tools of the third group in the second wave use Markdown, following the local-first principle. This way, the content of users, save for some syntactic differences such as block IDs and document metadata, is independent of the application. It can be viewed and manipulated from one application, say Obsidian, and then the user can continue editing with Logseq or Foam to use a specific feature. This is already a case of data-application decoupling and users having full control over their data. Logseq went a step further, providing a way to query the data independent of the Logseq application, and to convert the graph to RDF. [49]
第三个迹象是从核心应用程序提供的功能转向扩展和插件。通常情况下，某些技术趋势可以从技术专业人员最初认为有用的东西中看出。可以推测，这种生态系统的发展方式可以预示一种总体趋势。以 VSCode 为例 [50]，那里的数据与应用程序分离，可以与其他编辑器一起使用，更重要的是，开发人员可以使用他们选择的工具进行协作——这种高度的自主性与 Git 提供的凝聚力相平衡。然后，许多功能——对某些人来说，是他们工作的关键功能——都是由扩展提供的。更重要的是，许多扩展无缝地使用其他扩展的功能。现在，随着 Obsidian、Logseq 和 Roam 等工具扩展的数量、复杂性和采用率的增长，也观察到了类似的情况。
And the third sign is the shift from functionalities provided by the core application toward extensions and plug-ins. As is often the case, certain technology trends can be seen from what first appears to be useful to the technology professionals. It can be speculated that the way such an ecosystem evolves can signal an overall trend. If we take, for example, VSCode, [50] the data there is decoupled from the application, can be used with another editor, and, more importantly, developers can collaborate using their tool of choice – a great autonomy that is balanced with the cohesion provided by Git. And then many functionalities – for some people, the crucial functionalities for their work – are provided by extensions. More importantly, many extensions seamlessly use the capabilities of others. A similar situation is now observed with the growth in number, sophistication, and adoption of extensions for tools such as Obsidian, Logseq, and Roam.
第四个迹象是 Block Protocol [51] 和 Noosphere [52] 等协议的发展，以及 Agora（Ivanec，本卷）和 Samepage [53] 等服务的出现，它们提供了使用独立 TfT 创建的内容的无缝集成。
A fourth sign is the development of protocols such as the Block Protocol [51] and Noosphere, [52] and the appearance of services such as Agora (Ivanec, this volume) and Samepage, [53] which provides seamless integration of content created using independent TfTs.
第五个标志（或者更确切地说是推动因素）是各种去中心化技术的进步，例如内容寻址（Benet，2014 年）、分布式账本（Zichichi 等人，2020 年）、去中心化标识符（去中心化标识符 (DID) v1.0，2022 年），更具体地说是首次尝试在 IPFS 上使用 Solid（Parrillo 和 Tschudin，2021 年），以及最近的 Noosphere，被描述为“思想协议”和“基于 IPFS 的全球知识图谱”（Brander，2022 年）。
A fifth sign, or rather enabler, is the progress with various decentralized technologies such as content-addressing (Benet, 2014), distributed ledgers (Zichichi et al., 2020), decentralized identifiers (Decentralized Identifiers (DIDs) v1.0, 2022), and more concretely the first attempt of using Solid over IPFS (Parrillo & Tschudin, 2021), and recently Noosphere, described as a “protocol for thought,” and “worldwide knowledge graph on top of IPFS” (Brander, 2022).
结论
Conclusion
知识图谱正在经历一个完整的循环。从概念上讲，它们似乎解决了在个人层面（memex、Zettelkasten）组织信息的问题，然后在实践中在组织层面（第一个超文本系统）解决，之后是在全球层面（WWW，然后是 LinkedData 和 Google 知识图谱），然后回到组织层面（企业知识图谱），现在又回到个人层面（PKG）。
Knowledge Graphs are coming full circle. Conceptually they appeared to solve the problem of organizing information at a personal level (memex, Zettelkasten), then in practice at the organizational level (the first Hypertext systems), after that at global (WWW, and then LinkedData and Google Knowledge Graph), and then back to organizational (Enterprise Knowledge Graphs) and now back to personal (PKG).
如今，OKG、EKG 和 PKG 共存，潜在的协同效应巨大。OKG 和 PKG 可以相互丰富。第一批 EKG 的成功可以促进 PKG 的使用，这反过来可以帮助人们开始以图形方式思考，并更频繁地转向知识图谱来应对组织内部和外部的数据相关挑战。
Today OKG, EKG, and PKG coexist, and the potential synergy is enormous. OKGs and PKGs can enrich each other. The success of the first EKGs can boost the use of PKGs, which in turn can help people start thinking in graphs and turn to knowledge graphs more often to tackle data-related challenges inside and outside organizations.
在某种程度上，每个人都已经在使用 PKG，但尽管大多数节点都是数字的，但大多数边缘都是精神的。这就是我们过去为数字世界提供凝聚力的方式，但最近我们已开始将这种凝聚力外包给 PKG 应用程序。接下来，凝聚力必须由技术和协议提供，允许控制我们的数据，同时自由选择和组合应用程序、功能和基础设施。这将允许更多的创新，从而提高我们图形扩展思维解决复杂问题的能力。
In a way, everybody is using a PKG already, but while most of the nodes are digital, most of the edges are mental. This is how we used to provide cohesion to our digital world, but recently we have started outsourcing that cohesion to PKG applications. Next, cohesion has to be provided by technologies and protocols, allowing control over our data and, at the same time, freedom to choose and combine applications, capabilities, and infrastructure. This will allow more innovation that will increase the power of our graph-extended minds to solve complex problems.
笔记
Notes
[1] 有时它能让你更多地了解“提供定义的人”（Foerster & Poerksen，2002）。
[1] Sometimes it lets you know more “about the person supplying the definition” (Foerster & Poerksen, 2002).
[2] 在计算机科学和信息科学中，本体被定义为“概念化的明确规范”（Gruber，1993 年）。它通常包括实体类型（类）、关系类型（属性）的表示、正式命名和定义，以及定义本体术语在一个或多个论域中的预期含义的逻辑公理或规则。与通常特定于特定环境、应用程序或组织的数据模型不同，本体可以被许多应用程序和组织使用，并且可以在一个环境中创建并在另一个环境中使用。
[2] In computer science and information science, ontology is defined as “an explicit specification of a conceptualization” (Gruber, 1993). It usually encompasses the representation, formal naming and definition of entity types (classes), relationship types (properties) and logical axioms or rules that define the consequences of the intended meaning of the ontology terms within one or more domains of discourse. Unlike data models, which are often specific for a certain context, application or organization, ontologies can be used by many applications and organizations, and can be created in one context and used in another.
[3] 在 RDF 中，“属性”一词既用于表示两个事物之间的关系，换句话说，表示两个节点之间的边的含义，也用于将事物链接到简单的文字值，例如数字、日期或字符串。
[3] In RDF, the term “property” is used both for relation between two things, in other words for the meaning of an edge between two nodes, and to link a thing to a simple literal value, such as number, date or a string.
[4] 但有趣的是，skos:definition 是 skos:note 的子属性。我选择将其解释为对定义地位的合理降级。功能上的好处是，当查询 skos:node 的所有值时，换句话说，查询用 skos:node 标识的有向边指向的所有节点，除了定义之外，还可能提供示例、范围和历史记录注释。它们可以通过 SPARQL 引擎或推理引擎检索，推理引擎将应用子属性公理并将 skos:note 的子属性的值作为 skos:note 的推断值。
[4] But interestingly skos:definition is a subproperty of skos:note. I choose to interpret this as rightful demoting of the status of definition. The functional benefit is that when one queries for all values of skos:node, in other words, all nodes that the directed edge identified with skos:node points to, this will potentially provide, besides definition, also examples, scope and history notes. They can be retrieved either by a SPARQL engine or by an inference engine that will apply the sub-property axiom and will bring the values of the subproperties of skos:note as inferred values of skos:note.
[5] 使用 SKOS 和任何其他 RDFS 或 OWL 本体可以将资源的身份（通过 URI 表示）与其人类可读的标签分离。可以有多个标签（每种语言一个），通过相同的属性链接到相同的 URI，而属性只是另一个 URI。当每种语言有多个标签时，最好对不同类型的标签使用不同的属性，例如 skos:altLabel 和 skos:hiddenLabel 的情况。
[5] The use of SKOS and any other RDFS or OWL ontology allows decoupling between the identity of the resource, represented through an URI, and its human-readable label. There can be many labels, one per language, linked to the same URI thought the same property, which is just another URI. When there is more than one label per language, it is better to use a different properties for the different types of labels like it is the case with skos:altLabel and skos:hiddenLabel.
[6] 在关系和面向对象范式中，需要提前定义模式。要将某些东西存储在 SQL 数据库中，需要为该类型的事物创建一个表。要创建新对象，需要有一个新的类。在知识图谱中，事物的类型与存储机制无关。它可以稍后添加，并且可以不止一个。在 RDF 图的情况下，类型只是图中的另一个节点。推迟定义的可能性同样适用于定义属性和关系。这将语义知识图谱与需要写入时模式或读取时模式的系统区分开来。
[6] Within relational and object-oriented paradigms, the schema needs to be defined in advance. To store something in an SQL database, a table needs to be created for that type of thing. To create a new object, there needs to be a new class. In knowledge graphs the type of a thing is not linked to the storing mechanism. It can be added later and can be more than one. In the case of RDF graphs, the type is just another node in the graph. The same possibility to postpone definitions applies for defining attributes and relations. This distinguishes semantic knowledge graphs from systems requiring schema-on-write or schema-on-read.
[7] 这里的“形式上”指的是对人类和机器来说都没有歧义的定义。
[7] Here “formally” refers to definitions that are unambiguous for both humans and machines.
[8] 据说，当时人们周末很流行一项活动，就是尝试走过柯尼斯堡七座连接河流岛屿的桥梁，每座桥只能过一次。欧拉用数学方法证明了这个任务是不可能完成的，从而为图论奠定了基础，预示了拓扑学的发展，也破坏了柯尼斯堡居民周日散步的乐趣。
[8] The story goes that it was a popular weekend activity to try to walk over every one of Königsberg’s seven bridges, which linked islands in the river, crossing each bridge only once. Euler showed mathematically that this task was not possible, thus laying the foundation for graph theory, anticipating topology and spoiling Sunday walks for the good people of Königsberg.
[9] 结构耦合是“两个（或多个）系统之间反复相互作用的结果，导致其结构一致”（Maturana & Varela，1992）。
[9] Structural coupling is the “a history of recurrent interactions leading to the structural congruence between two (or more) systems” (Maturana & Varela, 1992).
[10] 卢曼认为，社会系统在其封闭的内部通信网络的基础上进行繁殖。科学就是这样一个社会系统，它通过内部的真/假代码运作。整个经济是一个不同的社会子系统，每个组织也是如此。将科学应用于工业是跨越系统边界进行交流的问题。这种交流只是一种扰动，可以根据自己的代码和参考触发科学之外系统的内部交流。因此，它在科学之外的任何地方都不可避免地会被误解。然而，科学交流在工业中可能具有相关性。在这些情况下，误解是有益的。
[10] According to Luhmann, social systems reproduce on the basis of their closed network of internal communications. Science is one such social system which works by its internal true/false code. The economy as a whole is a different societal subsystem, and so is every organization. Applying science in industry is a matter of communicating across a system boundary. Such communications are only a perturbation that can trigger internal communication in systems outside science according to their own code and references. As such, it is inevitably misunderstood anywhere outside science. Yet, scientific communication can have relevance in the industry. In those cases, the misunderstanding is productive.
[11] 如果我们询问（发送查询）一个图谱并得到一个很好的答案，我们可以说这个图谱知道答案，但那将是一个拟人化的隐喻。这种拟人化模式不仅在知识图谱中很常见。当人们谈论对自然语言提示作出反应的大型语言模型时，就会出现这种情况。但在那里，答案——尽管它可能令人印象深刻——甚至不是信息检索的问题。它只是答案的生成预测，“绕过”了语义、理解和常识。
[11] If we ask (send a query to) a graph and we get a good answer, we can say that graph knows the answer but that will be an anthropomorphic metaphor. Such an anthropomorphic pattern is common not only for knowledge graphs. It appears when people talk about large language models that react on prompts in natural language. But there the answer – as impressive as it may be – is not even a matter of information retrieval. It is just a generated predication of an answer, “bypassing” semantics, understanding and common sense.
[12] Google Knowledge Graph 是一个可公开访问的知识图谱，但它不像 DBpedia 和 Wikidata 那样透明。它的新 API 位于 https://cloud.google.com/enterprise-knowledge-graph/，这也是将其视为 OKG 和 EKG 之间的另一个原因。
[12] The Google Knowledge Graph is a publicly accessible knowledge graph but it is not as transparent as DBpedia and Wikidata. Its new API is on https://cloud.google.com/enterprise-knowledge-graph/ which is yet another reason to treat is somehow between OKG and EKG.
[13] 请参阅 https://www.ekgf.org
[13] See https://www.ekgf.org
[14] 例如，Seneviratne 等人将 Gyrard 的个性化健康知识图谱案例（Gyrard 等人，2018 年）称为“个人知识图谱”（Seneviratne 等人，2021 年）。
[14] For example, Seneviratne et al. refer to the Personalized Heath Knowledge Graph case of Gyrard (Gyrard et al., 2018) as “Personal Knowledge Graph” (Seneviratne et al., 2021).
[15] 我引用了 Manfred Kuehn 的译文（Kuehn，nd）。
[15] I used the translation of Manfred Kuehn (Kuehn, n.d.).
[16] 在这里，可能只有“真实”会让人感到不安。毕竟，描述迪士尼世界的图表应该是真正的知识图谱，而不是关于“现实世界”的。对此，作者之一 Juan Sequeda 在推特上做出了以下澄清
[16] Here, probably only “real” creates some unease. After all, a graph describing Disney World should be a proper knowledge graph and still not being about “the real world”. To which, Juan Sequeda, one of the authors, made the following clarification on Twitter
现实世界是指用户的话语领域。如果你关心的是“无形的东西”，那么它就是你的现实世界的一部分。
Real world refers to the domain of discourse of a user. If the “intangible thing” is what you care about, then it’s part of your real world.
然而，我并不确信形容词在定义中增加的价值，我建议删除它，至少在重新使用它来定义 PKG 时。
Yet, I’m not convinced of the value that adjective adds in the definition and I suggest dropping it, at least when it is reused to define a PKG.
[17] 做记号是区分的指示，而区分本身就是原子认知操作。乔治·斯宾塞·布朗认识到了这一点，并发展了指示演算（Spencer-Brown，1979），涉及生物学（F. Varela）、数学（L. Kauffman）、社会科学（N. Luhmann）和计算（W. Bricken）等学科。
[17] Making a mark is an indication of distinction which by itself is the atomic cognitive operation. This realization by George Spencer-Brown and the calculus of indications developed by him (Spencer-Brown, 1979) across disciples including biology (F. Varela), mathematics (L. Kauffman), social sciences (N. Luhmann) and computation (W. Bricken).
[18] 4E 代表着一种总体主张，即认知不仅存在于头脑中，而且是具体化的、嵌入的、实施的和扩展的。4E 认知可以看作是认知科学的第三个学派，但实际上是一场更广泛的运动，其中不同的学派，如扩展认知和实施主义，虽然共同反对计算范式的理论承诺，但也存在显著差异。
[18] 4E stands for the overarching claim that the cognition is not only in the head but is also embodied, embedded, enacted, and extended. The 4E cognition can be seen as the third school of cognitive science but is in fact a broader movement where different schools such as extended cognition and enactivism, although having in common their opposition of the theoretical commitments of computational paradigm, have significant differences.
[19] 总体而言，具身认知深受马图拉纳和瓦雷拉自创生理论的影响，他们将结构耦合定义为“导致两个（或多个）系统之间结构一致的反复相互作用的历史”（马图拉纳和瓦雷拉，1992）。
[19] The embodied cognition in general is heavily influenced by the theory of autopoiesis by Maturana and Varela, and they defined structural coupling as “history of recurrent interactions leading to the structural congruence between two (or more) systems” (Maturana & Varela, 1992).
[20] 除了生物和社会系统 (De Jaegher & Di Paolo, 2007; Paolo 等, 2018) 之外，情绪 (Colombetti, 2014) 和习惯 (Egbert & Cañamero, 2014) 似乎也是如此。
[20] Apart from biological and social systems (De Jaegher & Di Paolo, 2007; Paolo et al., 2018), this seem to be the case also for emotions (Colombetti, 2014) and habits (Egbert & Cañamero, 2014).
[21] 本卷的“尼克拉斯·卢曼的个人知识图谱”一章解释了这一现象。
[21] This phenomenon is explained in the chapter “The Personal Knowledge Graph of Niklas Luhmann” in this volume.
[22] 唯一实现一定程度互操作性的大众应用是电子邮件。我们可以使用一个应用程序编写电子邮件，收件人可以使用他们选择的邮件客户端打开、阅读和回复。代码开发人员的编辑器（即所谓的 IDE，如 VSCode）实现了更好的互操作性，但这只是个例，并不是大众市场的情况。
[22] The only mass application that achieved some level of interoperability was email. We can use one application to write an email and the addressee can open, read and reply to it with a mail client of their choice. An even better level of interoperability was achieved by editors for code developers, the so-called IDEs like VSCode but that is a specific and not a mass market case.
[23] 定向越野的定义是“通过一系列小步伐来缩小目标”（Teevan 等人，2004 年）。
[23] Orienteering is defined as “series of small steps were used to narrow in on the target” (Teevan et al., 2004).
[24] 早在数字技术出现之前，万尼瓦尔·布什就预见到了这个问题（见本节开头的引文）。
[24] Long before digital technologies appeared, Vannevar Bush envisaged this problem (see the quote starting this section).
[25] 例如，导航是搜索文件时的首选方法，但很少用于书签（O. Bergman 等人，2021 年），并且依靠电子邮件但效果不佳（Jones 等人，2014 年）。随着索引技术的进步，总体趋势是采用基于搜索的重新查找方法。
[25] For example, navigating was a preferred method when searching for files but rarely used for bookmarks (O. Bergman et al., 2021) and relied upon but not effective for emails (Jones et al., 2014). With the advancement of indexing technologies there is overall tendency towards search-based method of refinding.
[26] 最近，微软启用了文件系统、云存储和网络的综合搜索功能，当然，这是以使用其浏览器为代价的。
[26] Recently Microsoft enabled combined search in the files system, cloud storage and the web but, of course, at the price of using their browser.
[27] 某些浏览器虽然仍具有单独的书签搜索界面，但会将书签包含在搜索建议中。
[27] Some browsers, while they still have a separate search interface for bookmarks, include bookmarks in the search suggestions.
[28] 2020 年前，此类工具不到十几种，到 2022 年，则有 70 多种。其中一些工具将在未来几年内消失，“出生率”很可能会下降。
[28] From less than a dozen before 2020, in 2022 there were over 70 such tools. Some of them will not survive the next couple of years and the “birthrate” will most likely go down.
[29] 以前，我将第二波浪潮称为“PKG 应用”，但我仍然找到了一个合适的替代标签。
[29] Previously I referred to second wave as “PKG Apps” and I still find an appropriate alternative label.
[30] 另外还有针对用户、共享文档、共享文档和缓存的表格，但就用户创建内容而言，只有两个。
[30] There are also tables for users, sharing documents, sharing docs and caching, but in terms of user-created content, there are only two.
[31] 最近已停产。
[31] Recently discontinued.
[32] 在 Datalog/Datascript 中，它被称为“属性”，但它更像是有向图中的边，而不是 RDBMS 术语中的属性。
[32] It is called “attribute” in Datalog/Datascript, yet it works more like an egde in a directed graph, than like an attribute in RDBMS terms.
[33] 请参阅 https://github.com/kvistgaard/roamo
[33] See https://github.com/kvistgaard/roamo
[34] 请参阅 https://oasis-lab.gitbook.io/roamresearch-discourse-graph-extension/
[34] See https://oasis-lab.gitbook.io/roamresearch-discourse-graph-extension/
[35] 第二组中的工具 Logseq 也是基于 Markdown 的。
[35] Logseq, a tool from the second group is also Markdown-based.
[36]Cmap 是个例外，它更接近中心。它甚至有一个支持 OWL 的版本，但已停用。
[36]Cmap is an exception and rather closer to the center. It even had a OWL-aware version, discontinued.
[37] 请参阅《基本平衡》（Velitchkov，2020 年）第 3 章
[37] See chapter 3 of Essential Balances (Velitchkov, 2020)
[38] 例如，问题可以出现在捕获它的上下文中、会议中、问题日志和看板中、演示文稿中以及架构图中 (Velitchkov, 2021b)。
[38] For example, an issue can appear in the context where it was captured, a meeting, in the issues log and kanban board, in a presentation, and in an architecture diagram (Velitchkov, 2021b).
[39] 进化适应是指在进化过程中特征功能的转变。一个典型的例子是鸟类羽毛。它们最初的功能是调节体温。然后它们进化为捕捉昆虫、性展示，最后是飞行。类似的例子不仅在物种进化中可以看到，在技术进化中也可以看到。也许最受欢迎的例子是微波炉的发现。1945 年，雷神公司的珀西·斯宾塞口袋里的一块巧克力棒在暴露于主动雷达装置的微波时开始融化。
[39] Exaptation is a shift in a function of a trait during evolution. A classic example is bird feathers. Their initial function was temperature regulation. They then evolved for insect catching, sexual display, and finally for flight. Similar examples can be seen not only in the evolution of species but also in the evolution of technology. Maybe the most popular one is the discovery of the microwave oven. In 1945, a chocolate bar in the pocket of Percy Spencer at Raytheon started to melt when it was exposed to the microwaves from an active radar set.
[40] 在我的项目中，我开始将 GitHub 和 GitLab 视为图表，这得益于 GraphQL 的使用。这也是扩展适应的一个例子，因为我通过标签和工作流约定将问题用于各种项目项目，例如风险、原则、规则、决策等。
[40] In my projects I started treating GitHub and GitLab as graphs, stimulated by the use of GraphQL. This is also an example of exaptation since I’m using issues for various project items such as risks, principles, rules, decisions, and so on, through labelling and workflow conventions.
[41] 请参阅 Twitter 的 Twemex（https://tweethunter.io/）和 Thread Portal，以及维基百科的 Side Portal。
[41] See Twemex (https://tweethunter.io/) and Thread Portal for Twitter, and Side Portal for Wikipedia.
[42] 上世纪中叶，广播和电视的广泛普及也产生了类似的效果。它导致了信息的被动消费：
[42] The widespread adoption of radio a television in the middle of the last century had a similar effect. It led to passive consumption of information:
电视观众、广播听众、杂志读者所面对的都是各种元素——从巧妙的修辞到精心挑选的数据和统计资料——使他们能够以最少的困难和努力“做出自己的决定”。但这些包装往往做得非常有效，以至于观众、听众或读者根本无法做出自己的决定。相反，他们会将包装好的意见插入自己的脑海，就像将一盘磁带插入录音机一样。然后，只要觉得合适，他就会按下按钮并“播放”该意见。他的表现还算不错，无需思考。然而，在信息消费者也是主要信息提供者的情况下，却出现了类似的被动性和普遍缺乏批判性思维的情况，这令人费解。
The viewer of television, the listener to radio, the reader of magazines, is presented with a whole complex of elements – all the way from ingenious rhetoric to carefully selected data and statistics – to make it easy for him to “make up his own mind” with the minimum of difficulty and effort. But the packaging is often done so effectively that the viewer, listener, or reader does not make up his own mind at all. Instead, he inserts a packaged opinion into his mind, somewhat like inserting a cassette into a cassette player. He then pushes a button and “plays back” the opinion whenever it seems appropriate to do so. He has performed acceptably without having had to think. It is puzzling, however, to observe similar passivity, wide-spread lack of critical thinking in situations where the information consumers are also the main information providers.
（Adler 和 Doren，1972）
(Adler & Doren, 1972)
[43] 这种中心化有多种表现形式。在网络上，这是一种“垄断需求，使供应商品化”的新形式（Brander，2021），总体而言是一种新的资本主义形式（Zuboff，2019）。与此同时，基础设施服务、云平台和企业应用服务也都趋于中心化，这导致供应商锁定加剧，由于“单一”故障点设计，漏洞风险更高。
[43] This centralization has several manifestations. In the web, it’s a new form of “monopoly on demand to commodify supply” (Brander, 2021) and overall a new form of capitalism (Zuboff, 2019). In parallel, there is centralization of infrastructure services, the cloud platforms, and centralization of corporate application services, leading to increased vendor lock-in and higher vulnerability due to the “single”-point-of-failure design.
[44] Facebook 和谷歌总部之间的距离为 11 公里（7 英里）。
[44] 11km (7 miles) is the distance between the Facebook and Google headquarters.
[45] 从历史上看，DNS就是一个例子，但它也有去中心化的替代方案，例如命名数据网络，声称具有实质性的优势。
[45] An example is DNS, historically, but it also has decentralized alternatives, such as the named data networking, claiming substantial advantages.
[46] https://getdweb.net/
[46] https://getdweb.net/
[47] 请参阅https://www.go-fair.org/fair-principles/
[47] See https://www.go-fair.org/fair-principles/
[48] 解耦是知识管理的一个进化趋势。在文艺复兴时期，寻求更好的知识组织方法的学者们意识到，世界的表征不需要反映世界，知识的顺序也不需要反映知识 (Cevolini, 2018)。在知识与世界解耦、索引与知识解耦之后，下一个带来开放式探索的解耦创新是用免费索引卡代替装订的普通书籍。近年来，语义知识图谱再次成为传统关系数据库系统中紧密耦合的结构和语义解耦的问题（所谓的“写入时模式”）。
[48] Decoupling is an evolutionary trend in knowledge management. During the Renaissance scholars looking for better methods for knowledge organization realized that the representation of the world does not need to mirror the world and the order of knowledge does not need to mirror the knowledge (Cevolini, 2018). After knowledge was decoupled from the would and the index was decoupled from knowledge, the next decoupling innovation bringing open-ended exploration was the replacement of bound commonplace books with free index cards. In more recent years semantic knowledge graphs are again a matter of decoupling structure and semantics that are tightly coupled in traditional relational database systems (the so-called “schema-on-write”).
[49] 请参阅 https://github.com/logseq/rdf-export
[49] See https://github.com/logseq/rdf-export
[50] 在 Stack Overflow 于 2021 年进行的一项调查中，VSCode 排名第一，在超过 82,000 名调查参与者中，有 71% 的人选择了它，远远领先于其他所有语言（Stack Overflow 开发者调查 2021，2021 年）。
[50] In a survey by Stack Overflow in 2021, VSCode ranked as number one, selected by 71% of the over 82K participants in the survey, significantly ahead of all the rest (Stack Overflow Developer Survey 2021, 2021).
[51] https://github.com/blockprotocol/blockprotocol
[51] https://github.com/blockprotocol/blockprotocol
[52] 参见 Brander (2022) 和 https://github.com/subconsciousnetwork/noosphere
[52] See Brander (2022) and https://github.com/subconsciousnetwork/noosphere
[53] https://samepage.network
[53] https://samepage.network
第 2 章
Chapter 2
Niklas Luhmann 的个人知识图谱
The Personal Knowledge Graph of Niklas Luhmann
伊沃·维利奇科夫
介绍
Introduction
近年来，卢曼的 Zettelkasten 引起了前所未有的关注。十多年前，当我第一次了解它时，网上几乎没有任何资料，只有几篇博客文章和一些工具制作尝试。[1] 在接下来的几年里，这种情况并没有改变。这很遗憾，因为那时我试图将其中的想法融入我的笔记实践中。[2] 2016 年，卢曼 1981 年关于这个主题的唯一一篇论文的英文译本问世。[3] 然而，直到 2019 年，人们的兴趣仍然很低。
Luhmann’s Zettelkasten has been a magnet for an unprecedented level of interest in recent years. When I first learned about it over a decade ago, there was almost nothing available online save for a few blog posts and some tool-making attempts. [1] That didn’t change in the years following. That was a pity because it was then that I tried to incorporate ideas from it in my note-taking practice. [2] In 2016 an English translation of Luhmann’s only paper on the subject from 1981 appeared. [3] Yet, the interest remained low until 2019.
2019 年下半年，情况开始发生变化，起初很慢，但到了 2020 年，变化非常快。从那时起，人们的兴趣一直很高。[4] 如今，有成千上万篇关于 Luhmann 的 Zettelkasten 的文章和视频。
In the second half of 2019, things started to change, slowly at first, then very quickly in 2020. Since then, the interest has remained consistently high. [4] Today there are thousands of articles and videos about Luhmann’s Zettelkasten.
这波关注显然与网络笔记和研究工具的普及有关，现在被称为“思想工具”。
This wave of attention is clearly related to the proliferation of networked note-taking and research tools, now referred to as “Tools for Thought”.
大量出版物带来了各种观点、解释和实践。大多数 [5] 关于卢曼的 Zettelkasten 的文章和视频都解释了它是什么、它的机制，此外，许多文章和视频还展示了一些基于它的工具支持实践。这些资源实用且有用。他们的重点是借用和重新创造当代工具以增加对生产力和创造力的支持。然而，通过这样做，他们突出了并非源自卢曼的方法，并且在其他重要方面存在盲点，而这可能是真正的创新所在。
The extensive number of publications bring a range of perspectives, interpretations, and practices. Most [5] articles and videos on Luhmann’s Zettelkasten explain what it is, its mechanics, and many demonstrate, in addition, some tool-supported practices based on it. These resources are practical and useful. Their focus is on what to borrow and recreate in contemporary tools to increase their support for productivity and creativity. Yet by doing so, they highlight methods that don’t originate from Luhmann and have blind spots on other important aspects, which could be where the real innovation is.
其中一个盲点就是卢曼激进的系统思想，这体现在他创作期后半段的作品中。Zettelkasten 体现了卢曼向世界揭示的社会系统的自主性、自指性，并且从许多角度进行了如此精确的阐述。
One such blind spot is the radical systemic ideas of Luhmann, manifested in his work in the second half of the productive period. The Zettelkasten was an embodiment of the autonomous, self-referential nature of social systems Luhmann revealed to the world and elaborated with such precision and from so many angles.
社会系统理论的发展看起来就像一个由系统循环组成的分形。它的主要特征是自我指涉性，要求理论在描述内容时考虑自身的概率。也许一个不太明显的系统循环是，卢曼将他不断发展的世界观应用到他的工作方法中，这有助于他的工作方法的发展，而这又支持了他的社会系统理论的发展。如果我们放大“没有写作就不可能思考”（Luhmann，1981）[6]，我们可能会发现还有三个嵌套的循环。循环和自我指涉似乎不仅是他的理论的核心，也是他的方法的核心。
The development of social systems theory looks like a fractal of systemic loops. Its main characteristic, self-referentiality, requires that the theory account for its own probability as part of what it describes. Maybe a less obvious systemic loop is that Luhmann applied his evolving worldview to his working method, which contributed to the development of his working method, which supported the evolution of his social system theory. And behind “It is impossible to think without writing” (Luhmann, 1981), [6] if we zoom in, we may find three more nested loops. It seems circularity and self-reference were central not only in his theory but also in his method.
第二个盲点，也是密切相关的盲点，在于是什么让他的 Zettelkasten 与众不同。卢曼经常被称为该方法的发明者。这远非事实。然而，他的方法在很多方面都是独一无二的。其一是他的 Zettelkasten 可以看作是一个个人知识图谱。因此，它很可能是第一个个人知识图谱 (PKG)，或者至少是迄今为止使用时间最长（45 年）且输出无与伦比的个人知识图谱。这是本文的主张。但这一主张本身只有在它与知识图谱中的节点的关系中才有意义——它与这种笔记系统的自主性以及它与所有知识图谱的特征以及个人知识图谱特有的特征的关系。
The second, and closely related, blind spot is on what made his Zettelkasten different. Luhmann is often referred to as the inventor of the method. That’s far from the truth. Yet, his method was unique in many ways. One was that his Zettelkasten could be seen as a personal knowledge graph. As such, it is most likely the first personal knowledge graph (PKG) or at least one that was used for the longest period so far, 45 years, and with unmatched output. This is the claim of the current essay. But that claim by itself gets significance only in its relation – as it is with nodes in a knowledge graph – to the autonomous nature of this note-taking system and to how it is related to features of all knowledge graphs and those that are specific for personal knowledge graphs.
我们可以将 Luhmann 的 Zettelkasten 视为第一个个人知识图谱吗？如果是，为什么这很重要？由于它们处于截然不同的环境中，哪些一般知识图谱特征与他的作品一样重要？知识图谱如何可能拥有“自己的生命”？
Can we see Luhmann’s Zettelkasten as the first personal knowledge graph? If yes, why is this important? What general knowledge graph features are important in a similar way to his work as they are in very different contexts? How is it possible for a knowledge graph to have a “life of its own”?
本章试图回答这些问题并提出新的问题。本章的其余部分结构如下。
This chapter tries to answer these questions and provoke new ones. The rest of this chapter is structured as follows.
第一部分概述了他作品的规模和精髓。它并不试图总结卢曼的理论，而是概述了我们需要解释他的 Zettelkasten 的一些独特特征的一些核心概念。
The first section gives an idea of his work’s volume and essence. It doesn’t attempt to summarise Luhmann’s theory but outlines a few central concepts that we needed to explain some peculiar characteristics of his Zettelkasten.
第二部分展示了 Zettelkasten 为何以及如何有资格成为 PKG。接下来的部分将个人知识管理的演变视为一种解耦，以及这与使用知识图谱的主要好处之一有何关联。
The second section shows why and how the Zettelkasten qualifies as PKG. It is followed by a section viewing the evolution of personal knowledge management as one of decoupling and how this is related to one of the main benefits of using knowledge graphs.
第四部分重点关注 Luhmann 的 Zettelkasten 的另一个特定特性，即延迟分类和其他关联的能力，这也是知识图谱的一个特定优势。
The fourth section focuses on another specific feature of Luhmann’s Zettelkasten, the capability to delay classification and other associations, which is again a specific advantage of knowledge graphs.
第五部分讨论了自由探索的能力，这是知识图谱和 Luhmann 的 Zettelkasten 共同的特点。
The fifth section is about the affordances of free exploration, characteristic of the knowledge graph and of Luhmann’s Zettelkasten by virtue of being one.
最后一节试图回答索引卡系统如何能够拥有自己的生命力的问题。最令人信服的解释来自于它所促成的成果：卢曼的社会系统理论。
The last section tries to answer the question of how it is possible for an index card system to have a life of its own. The most convincing explanation comes from what it helped to produce: the social systems theory of Luhmann.
卢曼的作品和思想
Luhmann’s works and ideas
卢曼作品的规模和复杂性激发了人们对研究卢曼知识管理系统的兴趣。说卢曼是一位多产的作家，一点也不为过。他的档案列出了 1956 年至 1998 年间创作的 1,464 部作品。[7] 这个数字包括 70 多本出版的书籍。如此惊人的生产力对于那些熟悉他的作品并能欣赏其严谨性、连贯性和质量的人来说更加令人印象深刻。
What motivates the interest in studying Luhmann’s knowledge management system is the magnitude and sophistication of his body of work. To say that Luhmann was a prolific author would be an understatement. His archive lists 1,464 works created in the period 1956–1998. [7] This number includes over 70 published books. Such an astounding productivity is all the more impressive for those familiar with his work who can appreciate its rigor, coherence and quality.
对于理解为什么 Zettelkasten“独立于作者而拥有自己的生命”（Luhmann，1981）来说，卢曼创作时期的第二部分，即“自创生转向”（Seidl & Becker，2005）尤其有意义。在那个时期，从 20 世纪 80 年代初到他生命的尽头，卢曼发展了他的宏大的社会系统理论。1984 年出版的《社会系统》奠定了理论基础，并在 1997 年出版的最后一本书《社会理论》中应用于整个社会。在此期间，卢曼在不同的书中详细阐述了他的社会系统理论如何在社会的不同子系统中发挥作用，例如教育、艺术、科学、宗教、经济、政治和法律。
Of particular interest for understanding why the Zettelkasten “gets its own life, independent of its author” (Luhmann, 1981), is the second part of Luhmann’s productive period, the “autopoietic turn” (Seidl & Becker, 2005). In that period, from the early 1980s to the end of his life, Luhmann developed his grand social systems theory. The theoretical foundation was laid with Social Systems, published in 1984 and applied to society at large in the last book Theory of Society, published in 1997. In between, Luhmann elaborated in separate books on how his social system theory plays out in different subsystems of society, such as education, art, science, religion, economy, politics and law.
卢曼理论中的一个核心解释工具是自创生或自我生产的概念。它最初由 Humberto Maturana 和 Francisco Varela 在 20 世纪 70 年代初提出，用于解释生命和认知（Maturana & Varela，1972/1980）。自创生的概念已广泛应用于各个领域（Mingers，2013）。如今，它在生物学、控制论、认知科学和心灵哲学中具有极大影响力。
A central explanatory tool in Luhmann’s theory is the concept of autopoiesis or self-production. It was originally developed by Humberto Maturana and Francisco Varela in the early 1970s, to explain life and cognition (Maturana & Varela, 1972/1980). The concept of autopoiesis has been applied in a wide range of areas (Mingers, 2013). It is highly influential today in biology, cybernetics, cognitive science and philosophy of mind.
自创生系统的典型例子是活细胞。其所有组成部分均由其他组成部分在内部产生。细胞的半透膜也是由膜促成的内部化学反应产生的（没有半透膜，元素将分散，不会参与反应）。这样的系统是自指的，并且在操作上是封闭的。自指系统挑战了开放（与环境交换物质，例如汽车）和封闭（与环境交换能量，但不交换物质，例如计算机）之间的经典区别。它们由于在操作上是封闭的而更加开放。操作封闭是指依赖过程的封闭网络（自主系统）或生产过程的封闭网络（自创生系统），它已被用来解释自主性、个体化、能动性、生存力和认知。自指系统“更开放”，至少在生命系统的情况下，它们不仅可以与环境交换能量和物质，还可以进行持久的适应性变化，这给我们带来了相关的结构耦合概念。
The canonical example of an autopoietic system is the living cell. All its components are internally produced by other components. The cell’s semipermeable membrane is also created by internal chemical reactions which the membrane enables (without it the elements will disperse and not enter into reactions). Such a system is self-referential and operationally closed. Self-referential systems challenge the classical distinction between open (exchange matter with the environment, e.g., car) and closed (exchange energy, but not matter, with the environment, e.g., computer). They are more open by virtue of being operationally closed. Operational closure in the sense of a closed network of dependent processes (autonomous systems) or closed network for production processes (autopoietic systems) has been used to explain autonomy, individuation, agency, viability and cognition. Self-referential systems are “more open” in the sense that, at least in the case of living systems, they can exchange not only energy and matter with the environment but can also have lasting adaptive changes, which brings us to the related concept of structural coupling.
自创生系统在结构上与环境耦合。由于它们在操作上是封闭的，来自环境的扰动可以触发但不能决定系统的变化。这种变化由自创生系统的内部结构决定。如果你扔一块石头，你可以预测它的轨迹，但如果你扔一只活鸟，你就不能预测它的轨迹。如果你踢一个球，球的运动是由你的踢腿决定的，但如果你踢一只狗，反应是由狗决定的。同样，如果你把手指放进火焰里，你的“神经系统会构建热的感觉——火焰中不存在热量”（Seidl & Becker，2005）。
Autopoietic systems are structurally coupled with their environment. Since they are operationally closed, the perturbations coming from the environment can trigger but cannot determine the change in the systems. This change is determined by the internal structure of the autopoietic system. If you throw a stone, you can predict its trajectory, but you can’t if you throw a living bird. If you kick a ball, the movement of the ball is determined by your kick, but if you kick a dog, the reaction is determined by the dog. By the same token, if you put your finger in a flame, your “nervous system constructs the sensation of heat – the heat does not exist in the flame” (Seidl & Becker, 2005).
卢曼采用了自创生和结构耦合的概念，并将它们运用到他的系统理论中。他区分了三种自创生系统：生命系统、心理系统和社会系统。社会系统和心理系统在结构上是耦合的。心理系统由相互连接的思想网络复制而成。社会系统由通信网络复制而成。社会系统离不开心理系统，但它们的行为是由其内部结构、自指通信网络决定的，而不是由结构上与它们耦合的心理系统决定的。[8]
Luhmann took the concepts of autopoiesis and structural coupling and adapted them to his systems theory. He distinguished three kinds of autopoietic systems: living, psychic and social. Social and psychic systems are structurally coupled. Psychic systems are reproduced by a network of connected thoughts. Social systems are reproduced by a network of communications. Social systems cannot exist without psychic systems, but their behaviours are determined by their internal structure, by the self-referential network of communications and not by the psychic systems structurally coupled with them. [8]
因此，一个社会系统就是一张图，每一次交流都是一个节点，通过与另一个节点连接（指另一次交流）而构成。
As such, a social system is a graph, and each communication is a node, constituted as such by virtue of being connected (referring to another communication) to another node.
每一次交流都是一个事件，一旦产生便会消失，它是话语（如何、为什么）、信息（什么）和理解（话语与信息之间的区别）之间的区别的统一体。这就带来了第三个重要概念，即区别，它是所有观察的核心。卢曼采用了乔治·斯宾塞·布朗（Spencer-Brown，1979）的指示演算概念，并将其用作理解系统和子系统形成及其元素、过程和行为的主要解释工具。
Each communication is an event that disappears once it’s created and is created as a unity of distinctions between utterance (how, why), information (what) and understanding (the distinction between utterance and information). This brings the third important concept, that of distinction, which is central to all observations. Luhmann took the concept for the calculus of indications of George Spencer-Brown (Spencer-Brown, 1979) and applied it as the main explanatory tool for understanding system and subsystems formations and their elements, processes and behaviour.
有两个重要结论，它们与所有其他社会理论相悖，但同时也为卢曼的框架带来了主要的解释力。第一个反直觉的结论是，人不是社会系统的一部分，而是环境的一部分。第二个结论是，人不交流，只有交流才能交流。
There are two important conclusions, which are at odds with all other social theories and at the same time they bring the main explanatory power of Luhmann’s framework. The first counter-intuitive conclusion is that people are not part of the social systems but of their environment. The second is that people don’t communicate, only communications can communicate.
以上绝不是对卢曼思想的总结。其目的是在介绍一些基础概念和主张的同时，支持解释如何通过卢曼与 Zettelkasten 合作产生的创新，与这种工具的交互能够展现出协作的特性。这将是最后一节的主题。
The above was by no means a summary of Luhmann’s ideas. The purpose was, while giving a taste of some foundational concepts and claims, to support the explanation of how interaction with such a tool can exhibit properties of collaboration through the innovations that resulted from Luhmann’s collaboration with his Zettelkasten. And that will be the subject of the last section.
现在我们继续回答这个问题：我们是否可以将 Niklas Luhmann 的 Zettelkasten 视为个人知识图谱。如果是，这对其工作方式有何重要意义？这与在其他用例中带来类似好处的知识图谱的共同属性有何关系？
Now we move on to answer the question whether we can see Niklas Luhmann’s Zettelkasten as a personal knowledge graph. And if yes, in what way was this significant for how it worked? How is this related to common properties of knowledge graphs that bring similar benefits in other use cases?
Zettelkasten 作为 PKG
The Zettelkasten as PKG
事实和数据
Facts and figures
从 1952 年到 1997 年，卢曼用了 45 年间的时间开发并使用了他的 Zettelkasten。
Luhmann developed and used his Zettelkasten over a period of 45 years, between 1952 and 1997.
它有大约 90,000 张 A6 卡片存放在 27 个容器（24 个抽屉和 3 个纸板箱）中，分为 200 多个部分，包含 15,000 个书目参考资料和 3,200 个条目的关键字索引。
It has around 90,000 A6 cards stored in 27 containers (24 drawers and 3 cardboard boxes), split into a bit more than 200 sections, with 15,000 bibliographical references and a keyword index of 3,200 entries.
每张卡片在其部分中都有一个固定的位置，由其编号决定。这样可以轻松找到，并且可以在任何地方进行开放式增长。位置编号由部分编号构成，分隔符后是 [9] 一串（通常）交替的字母和数字（Schmidt，2018）。这样，与 1/1 直接相关的卡片将获得编号 1/1a。属于第 1 节的下一张卡片可以获得编号 1/1b，或者 - 如果它直接引用 1/1a - 1/1a1。同一部分中的另一张卡片，取决于它具体与什么相关，将在任何时候继续现有分支或开始新分支。
Each card has a fixed position in its section determined by its number. This allows easy finding and the possibility for open-ended growth anywhere. The position numbers are constructed by section number and, after a separator, [9] a sequence of (usually) alternating letters and numbers (Schmidt, 2018). This way, a card directly related to 1/1 will get the number 1/1a. The next card belonging to section 1, can get a number 1/1b or – if it directly refers to 1/1a – 1/1a1. Another card in the same section, depending on what it relates to specifically, will continue an existing branch or start a new branch at any point.
书目和关键字索引注释按字母顺序编号，两面书写并编号，例如 F、F2、F3 等。
Bibliographical and keyword index notes are numbered alphabetically, written on both sides and numbered, for example, F, F2, F3, and so on.
除了基于卡号的分支关系外，在源卡上还存在交叉链接，这些交叉链接以首选卡号标注，并作为被引用卡上的反向引用。Johannes Schmidt 估计有超过 50,000 个这样的关系。
Apart from the relations in a branch based on card numbers, there are cross-links noted on the source card by the number of the preferred card and as back-reference on the referred card. Johannes Schmidt estimated over 50,000 such relations.
个人知识图谱
Personal Knowledge Graph
Luhmann 的 Zettelkasten 是否符合个人知识图谱的资格？知识图谱由 Hogan 等人定义为
Does Luhmann’s Zettelkasten qualify as a personal knowledge graph? Knowledge graph is defined by Hogan et al. as
用于积累和传达现实世界知识的数据图，其节点代表感兴趣的实体，其边代表这些实体之间的关系。
a graph of data intended to accumulate and convey knowledge of the real world, whose nodes represent entities of interest and whose edges represent relations between these entities.
（Hogan 等人，2022 年）
(Hogan et al., 2022)
基于此，在上一章中，我对PKG提出了如下定义：
Based on that, in the previous chapter, I proposed the following definition for PKG:
用于积累和传达世界知识的数据图，其节点代表个人感兴趣的实体，其边代表这些实体之间的关系。
A graph of data intended to accumulate and convey knowledge of the world, whose nodes represent entities of personal interest and whose edges represent relations between these entities.
这可以视为三个相互关联的标准：
This can be viewed as three linked criteria:
1. 感兴趣的实体作为节点
1. Entities of interest as nodes
2. 这些实体之间的关系作为边
2. Relationships between these entities as edges
满足这两个条件将使信息结构有资格成为知识图谱。要成为个人知识图谱，我们需要一个额外的兴趣条件：
Satisfying these two conditions will qualify an information structure as a knowledge graph. To be a personal knowledge graph we need one additional condition on interest:
3. 个人兴趣
3. Personal interest
节点
Nodes
索引卡“zettel”是感兴趣的实体，因此它们自然是节点的嫌疑人。事实上，Zettelkasten 可以看作是一个具有多种类型节点和多种类型关系的图。关于节点，值得注意的是，卢曼理解持久性 [10] 标识符的重要性以及身份和语义之间的分离。事实上，他的标识符看起来非常像当代的不透明 URI，其中部分编号充当命名空间，使用正斜杠增强了这种相似性。
The index cards, the “zettels”, are the entities of interest, and so they would be the natural suspect for nodes. And indeed, the Zettelkasten can be seen as a graph with several types of nodes and several types of relations. Regarding the nodes, it’s notable that Luhmann understood the importance of persistent [10] identifiers and the separation between identity and semantics. In fact, his identifiers look very much like contemporary opaque URIs, with the section numbers serving as namespace, a semblance enhanced by the use of a forward slash.
图 2.1。 Zettelkasten 作为 zettels（卡片）图
Figure 2.1. Zettelkasten as a graph of zettels (cards)
但是将每张索引卡视为一个节点并不能揭示 Zettelkasten 网络的实际结构和密度，对书目和关键字卡也根本不起作用。将卡片看作它们的本质更有用：物理信息存储，其位置由其标识符决定。如果物理和逻辑组织是分开的，那么每个 zettel 都可以看作是一个媒介，上面有关于实体和关系的信息。然后我们可以区分一个逻辑 zettel，它具有相同的标识符，链接到它的内容，它本身可以是一个或多个节点，每个节点对应单独的段落甚至单词。[11] 一个物理 zettel 可以包含许多逻辑 zettel。这样的模型将对应于实际的粒度，考虑到集体和单一参考（Schmidt，2018）。
But treating each index card as a node would not reveal the actual structure and density of the Zettelkasten network and would not work at all for bibliographical and keyword cards. It is more useful to look at the cards as what they are: physical information storage with location determined by its identifier. If physical and logical organization are separated, then each zettel can be seen as a medium on which there is information about both entities and relations. Then we can distinguish a logical zettel, having the same identifier, linked to its content, which itself can be one or several nodes, each corresponding to separate paragraphs or even words. [11] One physical zettel can contain many logical zettels. Such a model would correspond to the actual granularity, having in mind the collective and single references (Schmidt, 2018).
在书目和关键字索引卡中，每个条目代表一个逻辑节点。
In bibliographical and keyword index cards, each entry represents a logical node.
边
Edges
在交叉引用的情况下，边由源节点和目标节点的编号标识。同一集合中的节点之间的关系将从卡号推断出来。第三种关系将以源卡中的特定位置作为源，以另一张卡作为目标，当该卡靠近时，仅通过大写字母来识别。[12]
An edge is identified by the numbers of the source and target node in case of cross reference. The relation between nodes in the same collection will be inferred from the card numbers. A third type of relationship will have as a source the particular place in the source card and, as a target, another card, which, when in close proximity, is identified only via a capital letter. [12]
提出一种合适的建模方法超出了本章的目标。它可以成为未来工作的主题。然而，SKOS [13] 的简单应用可以说明为什么 Luhmann 的 Zettelkasten 可以被视为知识图谱，并激发人们尝试更具表现力的建模。
Proposing an appropriate modelling approach is beyond the ambition of this chapter. It can be the subject of future work. Yet, a simple application of SKOS [13] can provide an illustration of why Luhmann’s Zettelkasten can be seen as a knowledge graph and inspire attempts for more expressive modelling.
以下是两个部分的示例，每个部分包含三张卡片。第一部分的一张卡片引用第二部分的两张卡片。这可以使用两个 SKOS 类（skos:Concept 和 skos:OrderedCollection）和三个属性（skos:memberList、skos:note 和 skos:note）来表示。
Here’s an example of two sections containing three cards each. One card of the first section is referring two cards of the second section. This can be represented using two SKOS classes, skos:Concept, and skos:OrderedCollection, and three properties: skos:memberList, skos:note, and skos:note.
图 2.2 使用 SKOS 的 Zettelkasten 的 RDF 图再现示例
Figure 2.2 Example RDF graph rendition of Zettelkasten using SKOS
此外，对于代表卡片的每个节点，可以有额外的元数据三元组，例如创建时间。将其用于关系将需要不同的建模方法或使用 RDF-star。
In addition, for each node representing a card, there can be additional triples for metadata, such as creation time. Having that for relations will require a different modelling approach or using RDF-star.
个人的
Personal
要成为 PKG，其实体以及其内容总体上应符合个人兴趣。由于 Luhmann 的 Zettelkasten 中的所有卡片都是根据他的个人研究兴趣创建的，因此满足了这一条件。
To be a PKG, the entities, and overall, its content should be of personal interest. Since Luhmann created all cards in his Zettelkasten based on his personal research interest, this condition is satisfied.
Balog 和 Kenter (2019) 将个人知识图谱定义为
Balog and Kenter (2019) define a personal knowledge graph as a
关于实体及其之间关系的结构化知识的来源，其中实体及其之间的关系具有个人而非一般的重要性。该图具有特定的“蜘蛛网”布局，其中图中的每个节点都连接到一个中心节点：用户。
source of structured knowledge about entities and the relation between them, where the entities and the relations between them are of personal, rather than general, importance. The graph has a particular “spiderweb” layout, where every node in the graph is connected to one central node: the user.
Zettelkasten 中的所有节点都与用户相连。Luhmann 的 Zettelkasten 的任何卡片上都没有关于这一事实的记录，但在图 2.2 中的电子版本中，可以使用更新的语句记录用户与每个节点的交互，并使用例如流行的 dcterms:created 和 dcterms:modified，图表确实会有一个蜘蛛网布局。有趣的是，Luhmann 本人将他的 Zettelkasten 称为“spedireweb”（Cevolini，2018）。
All nodes in the Zettelkasten are connected to the user. There is no record of this fact on any card of the Luhmann’s Zettelkasten, but in an electronic version like the one in figure 2.2, the user interaction with each node can be recorded with updated statements and using e.g., the popular dcterms:created and dcterms:modified, the graph will indeed have a spiderweb layout. Interestingly, Luhmann himself referred to his Zettelkasten as “spedireweb” (Cevolini, 2018).
尽管卢曼并没有隐瞒他的知识管理实践，并向其他人展示了它，但它似乎只对其创造者有用。卢曼在 zettel 9/8.3 上反映了这一点：[14]
Although Luhmann wasn’t secretive about his knowledge management practice and showed it to other people, it seems it was only of use to its creator. Luhmann reflected this on zettel 9/8.3: [14]
人们来到这里，就像在色情电影中一样，他们看到了一切，但仅此而已；因此，他们失望而归。
People come, they see everything and nothing more than that, just like in porn movies; consequently, they leave disappointed.
而这并非卢曼的 Zettelkasten 所特有的：“相同的摘录不能促使每个人用相同的推理回忆起相同的意义联想”（Cevolini，2018）。这种认识并不新鲜。在同一篇论文中，Cevolini 指出，弗朗西斯·培根“对学者之间分享摘录的可能性持怀疑态度”。
And that is not specific to Luhmann’s Zettelkasten: “same excerpts cannot prompt everyone to recall the same meaning associations with the same reasoning” (Cevolini, 2018). Such realization is not new. In the same paper Cevolini points out that Francis Bacon was “skeptical about the possibility that excerpts might be shared among scholars”.
解耦
Decoupling
新笔记放在哪里并不重要，如果有多种可能性，我们可以按照自己的意愿解决问题，只需通过链接记录连接即可。
It is less important where we place a new note. If there are several possibilities, we can solve the problem as we wish and just record the connection by a link.
尼克拉斯·卢曼
Niklas Luhmann
个人知识管理和知识表示的历史中存在一种趋势，当代“解耦”概念很好地体现了这一趋势。在软件工程中，该术语与 20 世纪 60 年代和 70 年代的模块化编程有关，然后在本世纪初应用于服务架构。最近出现了将数据与应用程序解耦的运动和实践。[15] 但解耦过程可以追溯到更早。
There is a trend in the history of personal knowledge management and knowledge representation in general, that is well captured by the contemporary concept of “decoupling”. In software engineering, the term was associated with modular programming in the 1960s and 1970s, and then it was applied at the beginning of this century for service architecture. Recently there are movements and practices on decoupling data from applications. [15] But the process of decoupling can be traced much earlier.
脱钩带来灵活性和自由。文艺复兴时期更大的政治、艺术和学术自由与个人知识管理的创新相互影响，而个人知识管理的创新也以增加自由为特征。首先是将一些记忆能力转移到外部工具上，让大脑自由地解决问题和进行创造性任务。这是摘录的古老艺术及其信息技术——普通书籍的复兴。但在文艺复兴时期，它们的功能得到了扩展：普通书籍从记忆辅助工具演变为辅助记忆，从支持记忆的工具演变为支持遗忘的工具（Cevolini，2018）。这种转变伴随着几种脱钩。
Decoupling brings flexibility and freedom. The greater political, artistic, and academic freedoms of the Renaissance period had a reciprocal influence with innovations in personal knowledge management that were also characterized by increased freedoms. The first was to offload some memorizing capacity to external tools and free the mind for problem-solving and creative tasks. This was the revival of the old art of excerpting and its information technology, commonplace books. But during the Renaissance, there was an exaptation of their function: commonplace books evolved from memory aids to become secondary memories, from tools supporting memorizing to tools supporting forgetting (Cevolini, 2018). There are several decouplings associated with this shift.
首先，知识表示现在是按照其自身的逻辑而不是它所代表的世界的逻辑来组织的。第二个脱钩是“存储顺序不再与知识顺序一致”（Cevolini，2018）。但可能最重要的脱钩是将书页从普通书籍的物理装订中解放出来。我们不知道是谁开始了这种做法，但从著名学者的角度来看，瑞士医生和博物学家康拉德·格斯纳（Conrad Gessner）可能是 16 世纪第一个管理大量个人笔记的人（Sawday & Rhodes，2000）。17 世纪，托马斯·哈里森（Thomas Harrison）凭借他的“研究方舟”（Malcolm，2004）将这种做法提升到了另一个复杂的水平。脱钩的优势是增强的自主性，只有在通过增强凝聚力来平衡的情况下才能实现（Velitchkov，2020）。单独保存笔记是不够的，只有有了新技术或装订方法才能发挥作用，只是比书籍更灵活。在哈里森的发明中，平衡是由一个设计精良的设备提供的，这是一个小柜子，单独的笔记被挂在主题标题下的挂钩上。
First, knowledge representation was now organized according to its own logic and not that of the world it represented. The second decoupling was that “storing order no longer coincides with the order of knowledge” (Cevolini, 2018). But probably the most significant decoupling was freeing the pages from the physical binding of commonplace books. We don’t know who started this practice, but from the famous scholars, Conrad Gessner, a Swiss physician and naturalist, was probably the first to manage an extensive collection of individual notes in the 16th century (Sawday & Rhodes, 2000). This practice was raised to another level of sophistication by Thomas Harrison in the 17th century with his “Ark of Studies” (Malcolm, 2004). The advantage of decoupling, which is the increased autonomy, can only be realized if balanced by increased cohesion (Velitchkov, 2020). Keeping separate notes is not enough and can only work if there is a new technology or method of binding, just more flexible than that of books. In the case of Harrison’s invention, the balance was provided by a well-designed device, a small cabinet, where individual notes were attached on hooks under subject headings.
在接下来的几个世纪里，这种做法继续发展。没有证据表明卢曼意识到了这一点。但他通过保持笔记的原子性并应用允许内部增长的分支系统，提高了脱钩的水平。随着笔记实践的发展，他还进行了另一次重要的脱钩。第一阶段（1952-1962 年）的主题导向转变为问题导向（Schmidt，2018 年）。这可以从笔记的内容和顶级类别的大幅减少（从 108 个减少到 11 个）中看出。在“我们可以选择主题专业化的路线（例如关于政府责任的笔记），也可以选择开放组织的路线”中，卢曼选择了后者，因为“寻找将异质事物相互关联的问题的表述更有成效”（Luhmann，1981 年）。最近，Sönke Ahrens [16] 和 Andy Matuschak 也表达了同样的认识。[17]
Such practices continue to evolve in the following centuries. There is no evidence that Luhmann was aware of them. But he increased the level of decoupling by keeping the notes atomic and by applying his branching system that allowed inward growth. There is another important decoupling that he did as his note-taking practice evolved. The subject orientation in the first period, 1952–1962, shifted to a problem orientation (Schmidt, 2018). This can be seen both from the notes’ content and the great reduction of top-level categories, from 108 to 11. In “we can choose the route of thematic specialization (such as notes about governmental liability), or we can choose the route of an open organization”, Luhmann opted for the latter because “it is more fruitful to look for formulations of problems that relate heterogeneous things with each other” (Luhmann, 1981). This realization has been echoed recently by Sönke Ahrens [16] and Andy Matuschak. [17]
Luhmann 的 Zettelkasten 是一个知识系统，从 20 世纪 50 年代初开始，存储、结构和语义就被分离了。但个人、不同行业和社会都以自己的速度发展。Luhmann 所达到的实现，以及他以自己的方式构建信息的方式，几乎不为人所知，也没有以任何方式影响信息管理的发展。他所实现的分离，无论文件柜的物理限制如何，在没有这种限制的计算中都尚未实现。这可以用经济和技术因素来解释。首先，存储成本很高。其次，行业需求与个人知识管理的需求不同。最后，实现这一点的技术尚未开发出来。正如 Dave McComb 所说：
Luhmann’s Zettelkasten is a knowledge system where, from the early 1950s, storage, structure and semantics were decoupled. But individual people, different industries, and societies develop at their own pace. The realization Luhmann reached, and implemented in his individual way of structuring information, was hardly known and did not influence the development of information management in any way. The decoupling he implemented, regardless of the physical constraints of his filing cabinet, was yet to be realized in computing that did not have such constraints. This can be explained by both economic and technological factors. First, the cost of storage was high. Second, the industry demands were different from those of personal knowledge management. And last, the technologies to enable this were yet to be developed. As Dave McComb put it:
过去二十年最深刻的进步之一就是能够将信息的含义与其结构分开。在传统系统中，含义既与数据结构紧密相关，又无法从数据结构中发现。
One of the most profound advancements in the last two decades has been the ability to separate the meaning of information from its structure. In a traditional system, meaning is both horribly bound up in the data structure, and at the same time, not discoverable from it.
这种灵活性是知识图谱相对于其他数据管理方式的主要优势之一。这种“空间”上的解耦使得时间上的解耦成为可能，这是下一节的主题。但在此之前，值得一提的是两种新的解耦趋势，它们密切相关，并且通常由知识图谱实现。一个是数据和应用程序的解耦，另一个是内容和主机的解耦。这两种趋势都高度相关，并且已经在 PKG 中得到体现，如上一章所示。
That flexibility is one of the main benefits of knowledge graphs over other ways of managing data. And such a decoupling “in space”, enables decoupling in time, which is the topic of the next section. But before that, it’s worth mentioning two new decoupling trends, which are closely associated and often enabled by knowledge graphs. One is the decoupling of data and applications, and the other is the decoupling of content and host. Both trends are highly relevant and already experienced in PKG, as shown in the previous chapter.
延迟模式
Delaying schema
一个基于内容的系统，就像一本书的大纲，意味着我们做出的决定会让我们在未来几十年内遵守一定的秩序！如果我们认为沟通系统和我们自己都有发展的能力，那么这必然会很快导致定位问题。
A system based on content, like the outline of a book, would mean that we make a decision that would bind us to a certain order for decades in advance! This necessarily leads very quickly to problems of placement, if we consider the system of communication and ourselves as capable of development.
尼克拉斯·卢曼
Niklas Luhmann
当笔记的位置和初始分类不能将其固定在记录思想网络中时，新知识和偶然发现就可以在语义空间中移动该笔记，而不会在物理空间中永久移动它[18]。卢曼非常清楚这种延迟的重要性，以及他的Zettelkasten的网络结构如何实现这种延迟。节点的价值来自它与其他节点的连接。约翰内斯·施密特解释说：
When the location and initial classification of a note does not fix its place in the network of recorded thoughts, this allows new knowledge and serendipity to move that note in the semantic space without moving it permanently [18] in the physical space. Luhmann realized all too well the importance of this delay and how it was enabled by the network structure of his Zettelkasten. The value of a node comes from its connections to other nodes. Johannes Schmidt explained:
[卢曼] 主要关心的不是在将笔记收录到收藏中之前将想法发展到最完善的程度；相反，他基于这样的假设：只有将笔记与其他笔记联系起来，才能决定笔记是否有用——因此（在许多情况下）这是一个未来要决定的事情：通过在后来汇编的新笔记的背景下重新阅读笔记，或者在调查的背景下，即使用卡片索引作为新想法和出版物的数据库。
[Luhmann’s] main concern was not to develop an idea to maximum sophistication before including the note into the collection; rather, he operated on the assumption that a decision on the usefulness of a note could only be made in relating it to the other notes – and therefore would (in many cases) be a matter to be decided in the future: by re-reading the note in the context of new notes compiled afterwards or in the context of an inquiry, i.e., in using the card index as a database for new thoughts and publications.
在传统数据库系统中，数据与模式 [19] 严格分离，同时又完全依赖于模式。除非数据模型已经定义了数据的位置 (Abiteboul, 1997)，否则无法存储任何数据。这称为写入时模式。知识图谱的主要优点之一是它们允许“维护者推迟模式的定义，从而允许数据及其范围以比关系设置中通常更灵活的方式发展，尤其是在捕获不完整知识时”(McComb, 2019)。Dave McComb 称之为“稍后模式”，并指出它比写入时模式和读取时模式都更有优势 (McComb, 2019)。我更喜欢称之为“已知模式”，以保持其既定形式，并强调模式会随着领域知识的发展而发展。
In traditional database systems, data is strictly separated from schema [19] and, at the same time, fully dependent on it. No data can be stored unless its place is already defined by the data model (Abiteboul, 1997). This is known as schema-on-write. One of the main benefits of knowledge graphs is that they allow “maintainers to postpone the definition of a schema, allowing the data – and its scope – to evolve in a more flexible manner than typically possible in a relational setting, particularly for capturing incomplete knowledge” (McComb, 2019). Dave McComb calls that “schema later”, pointing out its benefits over both schema-on-write and schema-on-read (McComb, 2019). I prefer to call it “schema-on-know” to keep it in the established form and emphasize that the schema develops along with the knowledge of the domain.
以允许延迟分类并保持对新分类开放的方式组织信息的价值不仅得到了个人知识管理实践和数据架构演变的证实，而且也得到了组织知识管理的证实。Max Boisot 认为，过早的编码和抽象可能会限制组织的探索和创新。应该允许知识有机地发展。
The value of organizing information in a way that allows delaying its classification and keeping it open for new classifications is confirmed not only by the individual knowledge management practices and the evolution of data architectures but also in organizational knowledge management. Max Boisot argued that premature codification and abstraction could limit exploration and innovation in organizations. Knowledge should be allowed to evolve organically.
回到知识图谱，最后一点是关于其“知识架构”质量。每个知识图谱都有推迟架构定义的灵活性，但对于个人知识图谱，由于不需要协调并且数据治理要求极低，因此这种灵活性要高得多。
Back to the knowledge graph, there is one last point to make on their “schema-on-know” quality. Every knowledge graph has the flexibility to postpone the definition of schema, but for personal knowledge graphs, since there is no coordination needed and the data governance requirements are minimal, this flexibility is much higher.
行走中形成的路径
Path made in walking
流浪者啊，没有路：
Wanderer, there is no path:
路是靠步行走出来的。
the path is made by walking.
安东尼奥·马查多，流浪者
Antonio Machado, Wanderer
在图书馆中，查询的答案是旅程的终点；而在图表中，答案则是起点。自由向任何方向导航的能力是知识图谱的共同属性，也是 Zettelkasten 图表属性所固有的。一个想法、一个查询或一次随机浏览可以将用户引导至一条笔记，而这条笔记又可以引导至相邻或远处的引用笔记，或引导至反向链接指向的笔记，从而实现对知识网络的无限制探索。在知识图谱中，这种浏览方式称为“跟着你的鼻子走”。但在大多数情况下，它仅限于浏览。[20] 在个人知识图谱中，例如 Luhmann 的 Zettelkasten，用户不仅可以在浏览图表时产生新的心理联想和见解，还可以将其中的一些具体化。这可能意味着创建新的笔记、新的连接或新的分类。甚至节点是什么，也由用户自行决定。正如卢曼所解释的那样，“我们可以将[笔记]连接到任何地方——甚至可以连接到连续文本中间的某个特定单词”。在这种情况下，那个特定的单词将成为一个节点。
In a library, the answer to a query is the end of the journey; in a graph, it’s the beginning. The ability to navigate freely in any direction is a common property of knowledge graphs and is also inherent in the graph properties of the Zettelkasten. An idea, a query, or a random browsing can lead users to a note, which can then lead to adjacent or distant referred notes or to a note pointed to by a backlink, allowing for unrestricted exploration of the knowledge network. In knowledge graphs, this manner of browsing is called “follow-your-nose”. But in most of them, it is restricted to browsing. [20] In a personal knowledge graph, such as Luhmann’s Zettelkasten, the user can not only make new mental associations and insights while walking the graph but also materialize some of them. This can mean creating new notes, new connections or new classifications. Even what a node is, is at the discretion of the user. As Luhmann explained, “we can connect [notes] anywhere – even to a particular word in the middle of a continuous text”. In such a case, that particular word will become a node.
实现新的联想和见解将增加下一次探索与上一次探索不同的理由，但即使在此期间没有做出任何改变，它也会有所不同。Cevolini 解释说，遵循“相同的搜索路线”，“可以在用户进行的新阅读或推测所推荐的新链接关系的背景下获得新的含义”（Cevolini，2018 年）。
Materializing new associations and insights will add to the reasons the next exploration is different than the previous, but it will be different even if no changes are made in the meantime. Following “the same searching route,” Cevolini explains, “can obtain new meanings against the background of new linking relationships recommended by new readings or speculations performed by the user” (Cevolini, 2018).
在两个不同的探索会话中，用户可以按照不同的路径到达相同的节点，并且基于相同或不同的初始查询。在当代数据管理中，知识图谱的这一特性不仅使它们有别于 SQL，而且也使它们有别于其他 NoSQL 模型。图查询语言超越了标准关系操作（例如连接、并集和投影），并支持任意长度的路径遍历（Angles 等，2018 年）。
In two different exploration sessions, the user can reach the same node but following different paths, and based on the same or different initial queries. In contemporary data management, this feature of knowledge graphs sets them apart not only from SQL but other NoSQL models. Graph query languages go beyond standard relational operations such as joins, unions and projections, and support path traversal of arbitrary length (Angles et al., 2018).
在认知科学中，西班牙诗人安东尼奥·马查多的开篇引语启发了“在行走中铺就道路”这一隐喻。它抓住了这样一个中心思想：通过与世界的接触，我们创造了以前无法获得的行动和感知的新可能性（F. Varela，1987 年；FJ Varela 等人，1991 年）。没有预先设定的道路，“道路就是我们的脚步，在行走中铺就”（Thompson，2007 年）。
In cognitive science, the starting quote from the Spanish poet Antonio Machado inspired the metaphor “laying down a path in walking”. It captures the central idea that by engaging with the world, we create new possibilities for action and perception that were not available to us before (F. Varela, 1987; F. J. Varela et al., 1991). There is no pregiven path, “the path is our footsteps, laid down in walking” (Thompson, 2007).
同样的道理，用户在个人知识图谱中行走时，同时也在跟随和创造一条路径。一些提示来自当前的脚步，另一些来自旧的脚步，而一些新留下的足迹将影响未来行走的路径。
By the same token, a user walking through her personal knowledge graph is at the same time following and making a path. Some prompts come from the current steps, others from old ones, and some newly made footprints will influence the path made in future walks.
用户和 Zettelkasten 或个人知识图谱相互适应、共同进化。它们在结构上是耦合的。
The user and the Zettelkasten or the personal knowledge graph adapt to each other and co-evolve. They are structurally coupled.
结构耦合、自创生和操作闭合是卢曼社会系统理论和行为认知科学的共同基础。尽管从这一点来看，它们走上了不同的道路，但追随其中任何一条都可以帮助我们理解为什么像 Zettelkasten（或 PKG）这样的事物可以拥有自主性和自己的生命。
Structural coupling, autopoiesis and operational closure are shared foundations for both Luhmann’s social systems theory and the enactive cognitive science. Although from that point on they take different paths, following any of them can help us understand why a thing such a Zettelkasten (or a PKG) can have autonomy and a life of its own.
自己的生命
A life of its own
经过大量使用这种技术，一种次级记忆将会出现，一个我们可以不断与之交流的另一个自我。事实证明，它与我们自己的记忆相似，因为它没有一个完全构建的整体顺序，没有层次结构，而且肯定没有像书一样的线性结构。正因为如此，它才有自己的生命，独立于作者。
As a result of extensive work with this technique a kind of secondary memory will arise, an alter ego with whom we can constantly communicate. It proves to be similar to our own memory in that it does not have a thoroughly constructed order of its entirety, not hierarchy, and most certainly no linear structure like a book. Just because of this, it gets its own life, independent of its author.
尼克拉斯·卢曼
Niklas Luhmann
正如 Cevolini 所观察到的，在文艺复兴晚期，“从记忆辅助工具到二级记忆的转变在语义层面上也是可察觉的”（Cevolini，2018）。学习方舟的发明者托马斯·哈里森 (Thomas Harrison) 将他的设备称为“机器”，而不是书籍或图书馆。在随后的几个世纪里，许多机器被发明出来，并引发了理论和分类。分类的一个重要标准是它们的输入和输出之间的关系。当存在一对一关系时，机器是“平凡的”。输出仅取决于输入。当输出取决于之前的输出时，机器就是“非平凡的”（von Foerster，2003）。由于 Luhmann 的 Zettelkasten 能够根据其内部状态提供不同的输出，因此它至少是一台非平凡的机器。
As Cevolini observed, during the late Renaissance, the “transition from memory aids to secondary memories is also detectable on the level of semantics” (Cevolini, 2018). Thomas Harrison, the inventor of the Ark of Studies, called his device not a book or a library but a “machine”. In the centuries that followed, many machines were invented, and they attracted theories and classifications. An important criterion for classification is the relationship between their input and output. When there is a one-to-one relationship, the machine is “trivial”. The output depends only on the input. When the output depends on previous outputs, the machine is “non-trivial” (von Foerster, 2003). Since Luhmann’s Zettelkasten is able to give different outputs depending on its internal state, it is at least a non-trivial machine.
在平凡机器和非平凡机器之间的区别中，我们再次发现了语义上的转变。冯·福斯特称之为“不同的生物”（von Foerster，2003）。这是有充分理由的。几十年前，在 20 世纪 40 年代，与冯·福斯特在 Ratio 俱乐部合作的 Rosh Ashby 发明了 Homeostat，这是第一台具有自适应行为的机器（Ashby，1952/2014），随后又有其他发明，其中一些实际上有生物的名字，例如威廉·格雷·沃尔特创造的乌龟。
A semantic shift can be detected once again in the distinction between trivial and non-trivial machines. Von Foerster called “different creatures” (von Foerster, 2003). With a good reason. A few decades earlier, in the 1940s, Rosh Ashby, with whom von Foerster collaborated in the Ratio club, invented the Homeostat, the first machine with adaptive behaviour (Ashby, 1952/2014), followed by other inventions, some of them actually having the names of creatures, for examples the tortoises created by William Grey Walter.
不难看出，卢曼的 Zettelkasten 是一台非同寻常的机器，甚至可以自我调节。但如何将其解释为具有“自主性”、拥有“自己的生命”并充当“沟通伙伴”的东西（Luhmann，1981）？
It’s not difficult to see Luhmann’s Zettelkasten as a non-trivial machine and even as one capable of self-regulation. But how it can be explained as something with “autonomy” that gets a “life of its own” and acts as a “partner in communication” (Luhmann, 1981)?
一种解释路径是行走时遵循行为认知科学的指导。它的主要研究对象是生物个体及其与世界互动时产生的认知。但也有其他具有自我维持组织的系统表现出自主性和能动性。这些系统可能是瞬态系统，例如当两个人试图在狭窄的走廊中擦肩而过时出现的系统。或者它们可能是长期存在的，例如习惯（Egbert & Cañamero，2014）和情绪（Colombetti，2017）。
One explanatory path can be laid in walking with guidance by the enactive cognitive science. Its main objects of study are biological individuals and the cognition arising when they engage with the world. But there are other systems with self-sustained organization that exhibit autonomy and agency. These can be transient systems like the one emerging when two people try to pass each other in a narrow corridor. Or they can be long lasting, like habits (Egbert & Cañamero, 2014) and emotions (Colombetti, 2017).
有人声称，系统可以在与某些类别的软件工具（例如视频游戏）交互等情况下出现（Vahlo，2017）。
There are claims that systems can emerge in situations like interaction with some classes of software tools, for example, video games (Vahlo, 2017).
所有这些系统的共同点是，它们在操作上都是封闭的，在无机系统的情况下，它们在结构上与有机系统耦合 [21]。结构耦合和操作上封闭是控制论、社会系统理论和行为认知科学之间的共同概念。但后者给出了自己的定义：
What is common between all these systems is that they are operationally closed, and in cases they are inorganic, they are structurally coupled [21] with organic systems. Structurally coupled and operationally closed are shared concepts between cybernetics, social system theory and enactive cognitive science. But the latter gives its own definition:
如果对于构成系统一部分的任何给定过程 P，(1)我们可以在其促成条件中找到组成系统的其他过程，并且 (2)我们可以在系统中找到依赖于 P 的其他过程，则该系统在操作上是封闭的。
A system is operationally closed if, for any given process P that forms part of the system, (1) we can find among its enabling conditions other processes that make up the system, and (2) we can find other processes in the system that depend on P.
（Di Paolo 等人，2010 年）
(Di Paolo et al., 2010)
将其应用于 Luhmann 的 Zettelkasten，从卡片“仅从系统内的链接和反向链接网络中获得其质量”（Luhmann，1981）的角度，不仅有助于解释其自主性，而且还将其直接与知识图谱联系起来。
Applying this to Luhmann’s Zettelkasten from the perspective that a card “receives its quality only from the network of links and back-links within the system” (Luhmann, 1981) can help not only to explain its autonomy but also to relate it directly to being a knowledge graph.
这种尝试应用生成主义框架来理解 Zettelkasten 的逼真特性的尝试只是部分的，它使用了核心思想之一，即自主性。全面的分析应该包括生成框架的其他四个相互交织的元素：意义建构、体现、涌现和体验 (Di Paolo 等人，2010)。它们可以为更广泛地分析用户和 PKG 之间的交互提供见解，如上一章“图形扩展思维”部分所示。希望这些草图能够启发未来的工作。然而，也存在一些已知的局限性。生成认知科学还没有发展到可以解释更高级的认知功能，这在研究和开发思维工具时尤其令人感兴趣。
This attempt to apply an enactivist framework to get an understanding of lifelike properties of Zettelkasten is only partial, using one of the core ideas, autonomy. A comprehensive analysis should include the other four intertwined elements of the enactive framework: sense-making, embodiment, emergence, and experience (Di Paolo et al., 2010). They can be insightful for a wider analysis of the interaction between a user and a PKG, as shown in the section “The graph-extended mind” in the previous chapter. Hopefully, these sketches can inspire future work. Yet there are some known limitations. The enactive cognitive science has not advanced to explain higher cognitive functions, and that would be of particular interest when studying and developing tools for thought.
现在，让我们换一种方式，看看卢曼自己的想法是否能解释像一件充满纸片的家具这样无生命的东西如何拥有自己的生命。他关于 Zettelkasten 的唯一一篇论文是这样开头的：
Now, let’s take another path and see if Luhmann’s own ideas can explain how something as inanimate as a piece of furniture full of slips can have a life of its own. His only paper on Zettelkasten starts with:
接下来是一段实证社会学。它与我和其他人有关，也就是我的纸条盒。
What follows is a piece of empirical sociology. It concerns me and someone else, namely my slip box.
（卢曼，1981）
(Luhmann, 1981)
我们可以将 Zettelkasten 视为一个社会系统吗？应用主流观点，至少有一种方式可以给出肯定的答案：与 Zettelkasten 互动，用户处于与过去和未来的自我的社交网络中。但是，卢曼对社会系统的看法（在上面的第一部分中提供了一瞥）并不包括社会系统中的人。这样做将与社会系统是自创生的主要主张相矛盾。自创生系统在操作上是封闭的，但并非所有操作上封闭的系统都是自创生的。鉴于上面提供的 Di Paolo 的定义，在操作上封闭的系统中，关系应该是启用的。对于自创生系统，元素不应简单地启用，而应由系统中的其他元素产生。这就是为什么卢曼社会系统中的核心要素是通信，而不是人。社会系统通过自指通信网络而存在，产生边界和有意识的系统，这些系统使通信成为可能，并且在环境中与社会系统在结构上耦合。
Can we see the Zettelkasten as a social system? Applying a mainstream view, there is at least one way to give a positive answer: interacting with a Zettelkasten, a user is in a social network with their past and future selves. But Luhmann’s view of social systems, a glimpse of which was provided in the first section above, does not include people in the social system. Doing so will contradict the main claim that social systems are autopoietic. Autopoietic systems are operationally closed but not all operationally closed systems are autopoietic. Given the definition of Di Paolo provided above, in operationally closed systems, the relations should be enabling. For an autopoietic system, the elements should not simply enable but be produced by other elements in the system. That’s why the central elements in Luhmann’s social systems are communications, not people. A social system comes into existence through a self-referential network of communications, producing a boundary and conscious systems, which make communication possible, and are in the environment, structurally coupled with the social system.
当应用于卢曼的Zettelkasten时，通信网络并不由单据网络来表示。
When applied to Luhmann’s Zettelkasten, the network of communications is not represented by the network of slips.
滑箱式沟通只有在高度普遍化，即建立起交往关系时才会富有成效。而且它只有在评价时才会富有成效，因此与某个时间有关，并且在很大程度上是偶然的。
The communication with the slip box becomes fruitful only at a high level of generalization, namely that of establishing communicative relations of relations. And it becomes productive only at the moment of evaluation, and is thus bound to a certain time and is to a high degree accidental.
（卢曼，1981）
(Luhmann, 1981)
交流是没有持续时间的事件。因此，口误包含话语，话语只能作为信息和理解之间的区别的统一体而产生交流，并且它们只能在参考其他交流时才能产生社会体系。
Communications are events without duration. Slips, then, contain utterances, which produce communication only as a unity of distinction between information and understating, and they can produce a social system only in reference to other communication.
笔记或任何其他媒介都无法存储信息，它们只有在被阅读时才能“产生”信息（von Foerster，2003），这一观点与贝特森的“产生差异的差异”相符，因为它应该对某人产生差异。卢曼的观点与贝特森的观点相近，但在信息和含义之间做了额外的区分。一条信息只有在表达新内容时才包含信息。如果它重复已知的东西，含义保持不变，但信息价值就会丢失（Luhmann et al., 2018）。卢曼的观点受到乔治·斯宾塞-布朗的《指示》中微积分的召唤定律的影响（Spencer-Brown, 1979）。重新召唤就是呼叫。
Notes, or any other medium, cannot store information, they can “yield” information only when read (von Foerster, 2003), a view in line with Bateson’s “a difference that makes a difference”, since it should make a difference to someone. Luhmann’s view is close to that of Bateson but makes an additional distinction between information and meaning. A message can contain information only if it says something new. If it repeats something that is known, the meaning stays the same, but the information value is lost (Luhmann et al., 2018). Luhmann’s view is informed by the law of calling of the Calculus in Indication of George Spencer-Brown (Spencer-Brown, 1979). The re-call is to call.
考虑到这一点，“沟通的最基本前提之一是合作伙伴可以相互制造惊喜”（Luhmann，1981）这一说法具有双重含义。一个是简单地告知对方，用户和 Zettelkasten，另一个是关于 Zettelkasten 成为惊喜生成器的能力。这种情况的先决条件是“机会不是偶然而是系统地重现”的行为是图结构，其中“单个查询可以触发参考网络”（Cevolini，2018）。但鉴于话语和交流之间的区别，个人知识图谱作为一个活生生的社会系统，不是所代表的系统，而是如前所述存储在纸条中的节点和关系，而是在用户与物理系统交互过程中出现的系统。
With this in mind, the statement “one of the most basic presuppositions of communication is that the partners can mutually surprise each other” (Luhmann, 1981) gets double meaning. One is about simply informing each other, the user and the Zettelkasten, and the other is about the ability of a Zettelkasten to be a surprise generator. A pre-condition for this is a behaviour where “chances are reproduced not by chance but systematically” is the graph structure, where a “single query can trigger a web of references” (Cevolini, 2018). But in view of the distinction between utterance and communications, the personal knowledge graph, as a living, social system, is not the one represented but the nodes and relation stored in the slips, as described earlier, but one that emerges during user interaction with the physical one.
结论
Conclusion
知识图谱的创建是为了提供刚性数据结构所缺乏的灵活性。首先，他们试图使开放数据共享更有用，然后整合企业中的应用程序孤岛，直到最近才支持个人知识管理。但早在计算时代之前，个人知识管理就走上了提高灵活性和开放性的道路。技术的进化，就像它所伴随的人类进化一样，被快速发展的时期所打断[22]。文艺复兴时期就有这样的发展，最近的一次是尼克拉斯·卢曼的笔记实践。他的方法虽然不是使用索引卡的创新，但在与知识图谱带来的创新类似的方面是创新的。在本章中，我们展示了Zettelkasten可以看作是第一个个人知识图谱，无论是从结构还是相关优势来看。这些优势来自于将存储、结构和含义相互解耦，并将重点从实体转移到它们的关系上。这样的个人知识图谱不仅可以根据需要提供可靠的见解和意外发现，而且可以成为一个称职的研究伙伴。
Knowledge graphs were created to provide the flexibility that rigid data structures lack. First, they tried to make open data sharing more useful, then to integrate application silos in corporations, and only recently to support personal knowledge management. But personal knowledge management, way before the era of computing, was on a path to increased flexibility and open-endedness. The evolution of technology, just like the evolution of humans it is coupled with, is punctuated [22] by periods of rapid developments. There were such developments in the Renaissance, and a more recent one, the note-taking practice of Niklas Luhmann. His method, although not an innovation in its use of index cards, was innovative in ways similar to the innovation brought by knowledge graphs. In this chapter we have shown that the Zettelkasten can be seen as the first personal knowledge graph, both by its structure and the associated advantages. These advantages come from decoupling storage, structure and meaning from each other and shifting the focus from entities to their relationships. Such a personal knowledge graph does not just provide reliable insights and serendipity on demand but becomes a competent research partner.
致谢
Acknowledgements
本章是我和我的 PKG 合作的产物。
This chapter is a product of collaboration between me and my PKG.
笔记
Notes
[1] 请参阅https://web.archive.org/web/20080101035827/http://takingnotenow.blogspot.com/2007/12/faithful-electronic-version-of-luhmanns.html
[1] See https://web.archive.org/web/20080101035827/http://takingnotenow.blogspot.com/2007/12/faithful-electronic-version-of-luhmanns.html
[2] 有一系列博客文章，始于https://www.strategicstructures.com/wordpress/?p=1194
[2] There is a series of blog posts starting from https://www.strategicstructures.com/wordpress/?p=1194
[3] 请参阅 Niklas Luhmann 所著的《Communicating with Slip Boxes》（Kuehn 出版社，2016 年）
[3] See Communicating with Slip Boxes by Niklas Luhmann (Kuehn, 2016)
[4] Google 趋势证实了这一印象（请参阅 https://trends.google.com/trends/explore?q=Zettelkasten&date=2010-01-19%202023-02-19#TIMESERIES）。2020 年夏季的峰值本身可能很有趣，但当忽略该峰值时，趋势会变得更加清晰。从 2019 年到 2022 年，人们的兴趣似乎稳步增长，此后一直处于高位。
[4] Google trends confirm this impression (see https://trends.google.com/trends/explore?q=Zettelkasten&date=2010-01-19%202023-02-19#TIMESERIES). The peak in summer 2020 is probably interesting in itself, but the trend gets clearer when that peak is ignored. It seems there was steady growth of interest from 2019 to 2022, and then it’s been at a high level since.
[5] 值得注意的例外是本章多处引用的约翰内斯·施密特 (Johannes Schmidt) 和阿尔贝托·塞沃里尼 (Alberto Cevolini) 的出版物。
[5] Notable exceptions are the publications of Johannes Schmidt, and Alberto Cevolini cited in many places in this chapter.
[6] 本节以及来自 Luhmann's Kommunikation mit Zettelkästen (Luhmann, 1981) 的所有其他引文均使用 Manfred Kuehn (Kuehn, 2016) 的翻译。
[6] This and all other citations from the Luhmann’s Kommunikation mit Zettelkästen (Luhmann, 1981) use the translation of Manfred Kuehn (Kuehn, 2016).
[7] 该名单由 Ernst Lukas 编制。截至 2022 年 2 月，名单包含 552 部未发表的作品。详情请参阅 https://niklas-luhmann-archiv.de/person/werkverzeichnis/uebersicht
[7] The list was generated by Ernst Lukas. It includes 552 unpublished works as of February 2022. See details at https://niklas-luhmann-archiv.de/person/werkverzeichnis/uebersicht
[8] 这种社会系统观点甚至对卢曼时代不存在的现象也证明了其解释能力，例如算法的社会影响（Esposito，2022 年）。
[8] This view of social systems proved its explanatory power even on phenomena that did not exist during Luhmann’s time, such as the social impact of algorithms (Esposito, 2022).
[9] 直到 1936 年，也就是前 23,000 张卡片，分隔符都是逗号，之后是斜杠（Schmidt，2018）。
[9] Up to 1936, or the first 23,000 cards, the separator was a comma and after that a forward slash (Schmidt, 2018).
[10] 关于卡号，卢曼解释说“只要我们不改变这个数字，从而也不改变纸条的固定位置”就足够了（Luhmann，1981）。
[10] Regarding the card numbers, Luhmann explained that its sufficient “that we never change this number and thus the fixed place of the slip” (Luhmann, 1981).
[11] “我们可以把[笔记]连接到任何地方——甚至可以连接到连续文本中间的某个单词” (Luhmann, 1981)
[11] “we can connect [notes] anywhere – even to a particular word in the middle of a continuous text” (Luhmann, 1981)
[12] 例如，参见 zettel 57,4e7b1a10f5f6 中的参考文献 A、B 和 C https://niklas-luhmann-archiv.de/bestand/zettelkasten/zettel/ZK_1_NB_57-4e7b1a10f5f6_V
[12] As an example, see reference A, B and C in zettel 57,4e7b1a10f5f6 https://niklas-luhmann-archiv.de/bestand/zettelkasten/zettel/ZK_1_NB_57-4e7b1a10f5f6_V
[13] SKOS – 简单知识组织系统 https://www.w3.org/2004/02/skos/intro
[13] SKOS – Simple Knowledge Organization System https://www.w3.org/2004/02/skos/intro
[14] Cevolini 译（Cevolini，2018 年）。
[14] As translated by Cevolini (Cevolini, 2018).
[15] 请参阅 http://datacentricmanifesto.org
[15] See http://datacentricmanifesto.org
[16] “在旧系统中，问题是：我应该把这条笔记存储在哪个主题下？在新系统中，问题是：我想在什么情况下再次偶然发现它？”（Ahrens，2017）。
[16] “In the old system, the question is: Under which topic do I store this note? In the new system, the question is: In which context will I want to stumble upon it again?” (Ahrens, 2017).
[17] Andy Matuschak 提出了“常青笔记”的概念，这一概念目前在高级笔记记录者和数字园丁中很受欢迎。常青笔记与临时笔记不同，是为了在各个项目之间不断发展和发挥作用而编写的。其中一个原则是它们以概念为导向，“而不是以作者、书籍、事件、项目、主题等为导向”（Matuschak，日期不详）。
[17] Andy Matuschak introduced the concept of “evergreen notes”, which is now popular among advanced note-takers and digital gardeners. Evergreen notes, unlike transient notes, are written to evolve and work across projects. One of the principles is that they are concept-oriented, “rather than by author, book, event, project, topic, etc” (Matuschak, n.d.).
[18] 将单独的、面向问题的原子思维记录在单独的卡片上，使用户能够将这些卡片物理地放在一起，对它们进行洗牌，总的来说，以各种方式与它们进行交互。从这个意义上讲，将它们移动到物理空间很重要（比较哈金斯）以及第一章中提到的关于具身认知和扩展认知的类似工作），但在这种互动之后，在纸质 Zettelkasten 的情况下，它们需要回到它们的永久位置，以便将来能够快速可靠地再次找到它们。
[18] Having individual, problem-oriented, atomic thoughts captured on separate cards allows a user to also physically bring these cards together, shuffle them and, overall, interact with them in various ways. In that sense, moving them into the physical space is important (compare Hutchins) and similar work on embodied and extended cognition referred to in the first chapter) but after such interaction, in the case of paper Zettelkasten, they need to go back to their permanent locations so that they can be quickly and reliably found again in the future.
[19] 通过使用不同的形式主义，这种分离得到进一步强调。在 SQL 中，模式用数据定义语言 (DDL) 表示，数据用数据操作语言表示。
[19] This separation is further emphasized by using different formalisms. In SQL, the schema is expressed in data definition language (DDL), and data is expressed in data manipulation language.
[20] 也有例外，比如维基数据，但编辑仅限于数据，而不能修改架构。
[20] There are exceptions, such as Wikidata, but editing is restricted to the data, and the schema cannot be modified.
[21] 有论据支持将一些不耦合的自然无机系统（如飓风）视为自主系统的可能性（Virgo，2011）。
[21] There are arguments supporting the possibility of regarding some not-coupled natural inorganic systems, such as hurricanes, as autonomous (Virgo, 2011).
[22] 这是指进化生物学中的“间断平衡”理论。
[22] This refers to the theory of “punctuated equilibrium” in evolutionary biology.
第三章
Chapter 3
评估个人知识图谱工具的框架
A framework for evaluating Personal Knowledge Graph tools
奥梅斯·巴尔特斯和乔治·阿纳迪奥蒂斯
简介：从笔记记录到个人知识图谱
Introduction: From note-taking to Personal Knowledge Graphs
记笔记是一种永恒的习惯。虽然大多数人都有记笔记的习惯，但每个人做笔记的方式和程度并不相同。而且，并非所有的笔记都具有相同的用途。笔记可以是临时的待办事项清单、简单的提纲，也可以是深思熟虑的评论。它们可能参考现有工作，也可能发展成为自己的详尽文档。
Note-taking is a timeless practice. While most people are in the habit of making notes, not everyone does this in the same way or to the same extent. And not all notes serve the same purpose either. Notes can range from ephemeral to-do lists and simple outlines to thoughtful remarks. They may reference existing work or grow to become elaborate documents of their own.
随着时间的推移，结构化的笔记记录方法（例如 Zettelkasten（Cevolini，2018））已正式化并被更系统化的人群所采用。通常，同一群人不仅倾向于系统化地记录笔记，而且还会采用和整合诸如使用日历和任务管理等做法。借助软件工具，个人知识管理和个人生产力实践变得越来越紧密。无论是否是故意为之，事实仍然存在。
Over time, structured approaches to note-taking, such as Zettelkasten (Cevolini, 2018), have been formalized and adopted by people on the more systematic end of the spectrum. Usually, it’s the same cohort that tends to not only be systematic in its approach to note-taking, but to also adopt and integrate practices, such as the use of calendars and task management. Aided by software tools, personal knowledge management and personal productivity practices have become increasingly intertwined. Whether it’s by design or not, the fact remains.
随着时间的推移，人们开发了大量软件工具来协助人们做笔记。这些工具的发展方向各不相同。有些工具旨在通过提供广泛的功能成为用户个人环境的中心，甚至扩展到协作功能，从而有可能跨越鸿沟进入专业环境。相比之下，其他工具则刻意将事情做得尽可能简单。
Over time, a multitude of software tools have been developed to assist in the practice of note-taking. These tools have grown in different directions. Some tools aim to become the center of the user’s personal environment by offering a wide array of functionality, even extending to collaborative features, potentially crossing the chasm to professional settings. In contrast, other tools have deliberately aimed to keep things as simple as possible.
然而，这两个类别都被一个不那么新但最近重新引入并受到赞赏的范式所改变：图表。关于图表及其历史、属性、用途和好处，可以说很多。
However, both categories have been transformed by a not so new, yet newly reintroduced and appreciated paradigm: graphs. A lot can be said about graphs, their history, properties, use and benefits.
使用图作为建模和结构化数据的隐喻被称为知识图谱。知识图谱是一种通用且灵活的数据抽象，是一种将世界视为有意义的连接网络的方式（Hogan 等人，2022 年）。
Using graphs as a metaphor for modeling and structuring data goes by the name of knowledge graphs. Knowledge graphs are a universal and flexible data abstraction, and a way to see the world as a network of meaningful connections (Hogan et al., 2022).
各种形状和大小的知识图谱正在激增，一个由构建者和用户、研究人员和爱好者组成的充满活力的社区正在围绕它们形成。越来越多的组织意识到知识图谱是统一和集成其异构数据结构和模型的最佳方式。这是推动采用所谓的企业知识图谱的众多用例中的一个突出用例。
Knowledge graphs in all shapes and sizes are proliferating, and a vibrant community of builders and users, researchers and enthusiasts is shaping around them. More and more organizations realize that knowledge graphs are the best way to unify and integrate their heterogeneous data structures and models. This is a prominent use case among many others driving the adoption of so-called enterprise knowledge graphs.
在个人信息管理领域应用相同的隐喻和原则会产生我们所谓的个人知识图谱 (PKG)。企业知识图谱 (EKG) 是多年工作和蓬勃发展的软件、实践和共享知识生态系统的产物。然而，这些最多只能部分适用于个人使用。
Applying the same metaphors and principles in the personal information management domain results in what we call personal knowledge graphs (PKGs). Enterprise knowledge graphs (EKGs) come backed with years of work and a burgeoning ecosystem of software, practices and shared knowledge. At best, however, those may be partially applicable to use in a personal capacity.
相反，我们目睹的是个人知识图谱新生态系统的诞生。这个生态系统由一些工具组成，这些工具虽然不一定基于企业知识图谱方法，但提供的功能在某种程度上类似于其应用程序。有时，企业知识图谱原则会被有意识地采用和调整。然而，更多的时候，这些原则要么被忽视，要么被重新发现。
Instead, what we are witnessing is the birth of a new ecosystem for personal knowledge graphs. The ecosystem is comprised of tools that, although not necessarily being based on enterprise knowledge graph approaches, offer functionality which on some level resembles their applications. Sometimes, enterprise knowledge graph principles are adopted and adapted consciously. More often than not, however, those principles are either ignored or rediscovered.
定义个人知识图谱工具：双向链接、图形导航和可视化
Defining personal knowledge graph tools: Bidirectional links, graph navigation and visualization
对于个人知识图谱，一切都归结为几个简单的原则和实践，可以帮助用户在使用笔记工具时无缝构建他们的知识图谱：提升连接，创建双向链接以及以图形形式导航和可视化笔记。
For personal knowledge graphs, it all comes down to a couple of simple principles and practices that can aid users in seamlessly building their knowledge graphs while working with note-taking tools: elevating connections, creating bidirectional links, and navigating and visualizing notes as a graph.
我们将个人知识图谱工具定义为提供双向链接支持以及以图形形式导航和可视化笔记的工具。
We define personal knowledge graph tools as the ones that offer support for bidirectional links, as well as navigating and visualizing notes as a graph.
不可否认的是，这个定义相当宽泛，尤其是与企业知识图谱的许多定义（通常是学术上精确的）相比。这主要有两个原因。
This definition is admittedly a rather loose one, especially compared to the many – often academically precise – definitions provided for enterprise knowledge graphs. There are two main reasons for this.
首先，绝大多数个人知识图谱工具实际上并非基于知识图谱数据存储。相反，它们通常使用其他类型的后端，并通过中间层和用户界面创建图谱。
First, the vast majority of personal knowledge graph tools are not actually based on Knowledge Graph data stores. Rather, they typically use other types of backends and create their graphs via intermediate layers and the user interface.
其次，与基于企业知识图谱创建的应用程序相比，这些工具面向不同的受众。在大多数情况下，用户不会具备企业知识图谱技术和实践的背景，甚至不了解这些技术和实践。因此，这些工具也采用了更为轻松、以最终用户为中心的方法。
Second, these tools address a different audience compared to the applications created based on enterprise knowledge graphs. In most cases, users will not have a background in, or even awareness of, enterprise knowledge graph technology and practices. Therefore, the tools also adopt a more relaxed, end-user centric approach.
越来越多的组织意识到，如果不能充分利用“大数据”的优势，那么没有背景的“大数据”并不一定是好事。元数据和不同信息之间的联系可以提供背景信息。这可以极大地帮助从数据中获取价值，这就是为什么链接成为企业数据中的一等公民，企业知识图谱正在激增的原因。
The fact that “big data” without context is not necessarily a good thing if you can’t leverage it to your advantage is being realized by more and more organizations. Metadata and connections among different bits of information can provide context. This can greatly assist in getting value out of data, which is why links are becoming first-class citizens in enterprise data and enterprise knowledge graphs are proliferating.
由于如今每个人必须管理的数据量和复杂性与几十年前企业主和知识管理专业人员必须管理的数据相当，因此链接的重要性也得到了越来越多用户的重视。
As the volume and complexity of the data each person must manage today is comparable to what business owners and knowledge management professionals had to manage a few decades ago, the importance of links is hitting home with more and more users, too.
当然，在笔记应用中使用链接并不是什么新鲜事。使用富文本编辑器，人们一直能够链接到外部资源。然而，大多数笔记工具最近增加了使用内部链接的功能。内部链接是指向其他笔记甚至笔记部分（如块）的链接。在大多数工具中，这是通过将用作内部链接的文本括在“[[]]”中来实现的。
Of course, there is nothing new about using links in note-taking apps. Using rich text editors, people have always been able to link to external resources. What is a recent addition for most note-taking tools, however, is the ability to use internal links as well. Internal links are links that point to other notes, or even parts of notes, such as blocks. In most tools, this is done by enclosing the text meant to act as an internal link in “[[]]”.
通常，通过将文本标记为内部链接，可以查找现有注释，而链接到不存在的注释会为其创建一个存根。这种做法是从维基百科中采用的，因此这种内部链接称为维基链接。一些维基百科还支持所谓的反向链接。反向链接是节点的传入链接，也可用于反向导航。例如，如果节点 A 链接到节点 B，那么也可以创建、跟踪、列出和可视化从节点 B 到节点 A 的反向链接。
Typically, by marking the text as an internal link, a lookup to existing notes is offered, while linking to a nonexistent note creates a stub for it. This practice is adopted from wikis, hence internal links of this sort are called wikilinks. Some wikis also support so-called backlinks. Backlinks are incoming links for nodes that can also be used to navigate in the inverse direction. For example, if node A links to node B, then a backlink from node B to node A can be created, followed, listed and visualized too.
通过反向链接，用户可以有效地创建自己的个人 wiki。这种简单但功能强大的机制对节点创建和连接大有裨益。虽然我们在 Web 上所知道的典型链接是单向的，但笔记应用中的内部链接是双向的。内部链接之所以被称为反向链接，是因为它们可以双向导航。使用反向链接可以快速在工作区中创建类似网络的结构。
Through backlinks users effectively create their own personal wiki. This simple but powerful mechanism goes a long way towards node creation and connectivity. While typical links as we know them in the Web are unidirectional, internal links in note-taking apps are bidirectional. Internal links are called backlinks because they can be navigated in both directions. Using backlinks quickly results in the creation of a web-like structure in one’s workspace.
除了双向导航之外，反向链接还可以利用与图相关的属性和算法。可以显示具有中心重要性的节点和集群或相关节点，可以发现和遍历连接节点的路径，可以发出和执行查询，并且图可视化可以帮助导航和模式识别。尽管与上一代笔记工具相比，链接在 PKG 工具中占据着更核心的作用，但它们仍然不是一等公民。例如，它们的方向既没有显示也没有被利用。
Besides bidirectional navigation, backlinks make it possible to leverage properties and algorithms pertinent to graphs. Nodes of central importance and clusters or related nodes can be surfaced, paths connecting nodes can be discovered and traversed, queries can be issued and executed, and graph visualizations can help navigation and pattern recognition. Although links occupy a more central role in PKG tools compared to the previous generation of note-taking tools, they are still not first-class citizens. For example, their direction is neither shown nor leveraged.
以图形隐喻和基元为基础的笔记应用程序正逐渐成为 PKG 工具。它们涵盖从简单的大纲到功能丰富的软件，用于创建相互关联的个人知识库，包括规划和可视化功能。
Note-taking applications underpinned by the graph metaphor and primitives are coming into their own as PKG tools. They run the gamut from simple outliners to feature-rich software for creating interlinked personal knowledge bases, including planning and visualization capabilities.
关于此作品
About this work
在本章中，我们将开始分析 PKG 工具的特性和功能。我们的目标是帮助读者全面了解 PKG 工具的软件概况。我们还希望帮助读者了解哪些工具可能适合他们的需求。
In this chapter we embark on an analysis of PKG tool features and capabilities. Our goal is to help readers get a broad overview of the software landscape of PKG tools. We would also like to help readers understand what tools might be a good match for their requirements.
由于这是首次系统地研究 PKG 领域，我们觉得有必要介绍一下它的起源和方法。这项工作由 Omes Baltes 发起，是他 2022 年在波鸿鲁尔大学提交的硕士论文的一部分，由 Maribel Acosta 指导。一份概述探索 PKG 工具各个方面的方向的摘要已提交给本书的征稿。
As this is the first effort to systematically approach the PKG domain, we feel a few words on its origins and approach are in order. This effort was initiated by Omes Baltes as part of his MSc thesis submitted at the Ruhr University Bochum in 2022, supervised by Maribel Acosta. An abstract outlining the directions for exploring various aspects of PKG tools was submitted to this book’s call for submissions.
这份投稿经过评估，本书联合编辑、专业行业分析师 George Anadiotis 也参与其中。原始材料是根据 Omes 的经验和对 PKG 领域的理解而创建的。George 在分析和数据点方面提供了额外的维度，设计了一个分析框架，并提供了定义、背景和叙述。这项工作还受益于审稿人 Ivo Velitchkov 和 Alex Rink 的建议。
This submission was evaluated and George Anadiotis, co-editor of the book and a professional industry analyst, also offered his involvement. The original material was created based on Omes’s experience and understanding of the PKG domain. George provided additional dimensions in the analysis and data points, devised an analytical framework, and provided definitions, context and narrative. This work also benefited from recommendations provided by reviewers Ivo Velitchkov and Alex Rink.
虽然这项工作并不声称能与专门的领域分析报告相提并论，[1] 但据我们所知，它是个人知识图谱领域中的第一份此类报告。因此，它的贡献主要在于定义和设计 PKG 工具的分析框架。它在标准或评估工具方面并不详尽。随着该领域的发展和成熟，我们希望能够很快重新审视它并在此基础上继续努力。
While this work does not claim to be on par with dedicated domain analysis reports, [1] it is the first of its kind for the personal knowledge graph domain that we are aware of. As such, its contributions lay mostly in defining and devising an analytical framework for PKG tools. It is not exhaustive in terms of criteria or evaluated tools. As the domain grows and matures, we hope to be able to revisit it soon and build on this work.
我们首先根据笔记范式和我们在评估中考虑的不同功能和非功能方面介绍 PKG 工具的简单分类法。然后介绍分类和评估的方法方面。随后，我们根据我们定义的方法介绍和比较几种工具。最后，我们提供了采用建议和该领域的展望。
We start by introducing a simple taxonomy for PKG tools based on the note-taking paradigm and the different functional and nonfunctional aspects we consider in our evaluation. Then we introduce methodological aspects of our classification and evaluation. Consequently, we present and compare several tools based on the methodology we defined. We conclude by providing recommendations for adoption and an outlook for the domain.
PKG 工具分类法
PKG Tool Taxonomy
越来越多的笔记工具满足上述 PKG 类别的分类标准。可视为 PKG 编辑器的工具允许将其知识元素（实体、页面和块）相互链接。链接采用嵌入和双向链接的形式，并且应可导航，理想情况下也应可直观显示。
There is a growing number of note-taking tools that satisfy the above proposed criteria for classification in the PKG category. Tools that can be considered PKG editors allow interlinking their knowledge elements – entities, pages and blocks. The links take the form of embeds and bidirectional links and should be navigable, ideally also visually.
大多数工具都是基于文本的，但它们的方法各不相同。我们将根据编辑器 UI（用户界面）和 UX（用户体验）的主要功能将它们粗略地分为几类。我们研究的不同方法是：
Most tools are text-based but their approaches differ. We will roughly separate them into categories based on the main features of their editor UI (user interface) and UX (user experience). The different approaches we examine are:
项目符号大纲编辑器：大纲编辑器是一种文本编辑器，其中每个段落都是可折叠列表中的项目符号。这些列表通常可以通过拖放重新排列。在某些大纲编辑器中，项目符号还可以相互引用或作为独立页面打开。
Bullet outliners: Outliners are text editors where every paragraph is a bullet point in a foldable list. These lists can often be rearranged with drag and drop. In some outliners the bullets can also reference each other or be opened as standalone pages.
Markdown 编辑器：Markdown 编辑器是支持 Markdown 语法的文本编辑器。Markdown 文件中的特殊字符会显示为常见的文本格式，如标题、项目符号列表、链接等。
Markdown editors: Markdown editors are text editors that support the Markdown syntax. Special characters in Markdown files get displayed as commonly found text formats like headers, bullet lists, links, etc.
块界面：块界面就像是 Markdown 和项目符号大纲的扩展。块不是段落，而是项目符号列表，而是内容周围的特殊容器。[2] 这些块可以是 Markdown 中的元素，例如列表、标题或图像。
Block interfaces: Block interfaces are like an extension of both Markdown and bullet outliners. Instead of paragraphs being bulleted lists, blocks are special containers around content. [2] These blocks can be elements found in Markdown, like lists, headers or images.
这种分类背后的原因值得详细阐述。由于 Markdown 是一种格式化语言，因此也可以说所有 PKG 工具都属于以下两个类别之一：项目符号大纲或块界面。但是，Markdown 编辑器对纯文本进行操作这一事实很重要。
It is worth elaborating on the reasoning behind this categorization. Since Markdown is a formatting language, it could also be argued that all PKG tools belong to one of two categories – bullet outliners or block interfaces. However, the fact that Markdown editors operate on plain text is important.
Bullet outliner 不是文件编辑器。它们的 UI 和 UX 差异很大，以至于无法仅通过文本更改某些内容。在 Markdown 中，只需删除周围的语法即可完成更改（例如编辑链接或格式选项）。在 outliner 中，这些操作需要通过 UI 执行。
Bullet outliners are not file editors. Their UI and UX have diverged so significantly that it’s not possible to change certain things with just text. In Markdown, changes, such as editing a link or formatting options, can be done by simply removing the syntax around them. In outliners those actions would need to be performed via the UI.
块界面在这方面走得更远。例如，当链接粘贴到 UI 中并转换为书签块时，无法通过编辑文本或语法来更改该块。Markdown 编辑器和大纲/块界面之间的另一个重大区别是，您无法进行全文搜索或搜索和替换，因为它不是纯文本。
Block interfaces go even further in that direction. For example, when a link is pasted in the UI and transformed into a bookmark block, there is no way to change that block by just editing the text or syntax. Another big difference between Markdown editors and outliners/block interfaces is the fact that you can’t do full text search or do search and replace, because it is not plain text.
Markdown 编辑器和区块界面通常具有更高级的功能，例如任务规划、提醒、与在线服务集成、嵌入或书签。这些功能可以作为综合平台的一部分提供，也可以通过插件生态系统提供。
Markdown editors and block interfaces often have more advanced functionality, such as task planning, reminders, integration with online services, embeds or bookmarks. This may be offered either as part of a comprehensive platform or via a plugin ecosystem.
虽然这些类别之间的界限很明确，但 PKG 工具属于哪个类别却往往不清楚。在我们的分析中，我们将根据我们认为最能代表它们的主要类别对工具进行分类。如果某个工具可能属于多个类别，我们将这些类别标记为次要类别。
While the boundaries between these categories are clear, oftentimes which category a PKG tool belongs to is not. In our analysis, we will classify tools under the primary category we feel best characterizes them. Where a tool may belong in more than one category, we will mark those as secondary.
评估 PKG 工具的框架
A framework for evaluating PKG tools
我们展示和评估 PKG 工具前景的第一步是绘制地图。与任何领域的任何评估过程一样，要考虑的一些参数是功能性的，而另一些则是非功能性的，一些是特定于领域的，而另一些则是通用的。在这里，我们定义了我们认为重要的参数，以便评估 PKG 工具。值得澄清的是，我们专注于基于笔记记录方法的 PKG 工具。这些工具是 PKG 领域中充满活力的一部分，可以说是更重要的一部分，但它们也不能说明其全部。
The first step on our journey to present and evaluate the PKG tool landscape is to map the territory. Like any evaluation process for any domain, some of the parameters to consider are functional while others are nonfunctional, some are domain-specific, while others are generic. Here, we define the parameters we consider important in order to evaluate PKG tools. It is worth clarifying that we focus on PKG tools based on note-taking approaches. These tools are a vibrant part of the PKG landscape, arguably the bigger part too, but they do not account for its entirety either.
非功能参数
Nonfunctional parameters
PKG 工具基于数据运行，要么创建全新的笔记和文档，要么存储和扩充现有笔记和文档。因此，需要考虑的非功能性参数中很大一部分与数据方面直接相关。数据存储的位置和方式、数据是否可以跨模式同步以及数据在不同工具之间无缝交换的程度，都是具有重大影响的参数。
PKG tools function on top of data, either creating entirely new notes and documents or storing and augmenting existing ones. Therefore, a significant part of the nonfunctional parameters to consider are directly related to aspects having to do with data. Where and how data is stored, whether they can be synchronized across modalities, as well as the degree to which they can be seamlessly exchanged between different tools, are all parameters with significant ramifications.
我们还研究了许可证和定价的相关参数。这些参数可以以多种方式组合 - 专有工具与免费层、开源工具与付费层、不同的定价方案等等。同样，在考虑工具的生态系统时，开源并不一定意味着健康的生态系统，专有并不一定意味着缺乏生态系统。最后但并非最不重要的是，供应商的可信度很难客观衡量，但却非常重要。
We also examine the related parameters of license and pricing. These can be combined in a number of ways – proprietary tools with free tiers, open source tools with paid tiers, different pricing schemes, and so on. Similarly, when considering a tool’s ecosystem, being open source does not necessarily imply a healthy ecosystem, and being proprietary does not necessarily imply a lack of an ecosystem. Last but not least, vendor credibility is hard to measure objectively, but is incredibly important.
数据格式
Data format
如上所述，并非所有 PKG 工具都采用相同的方法。这意味着不同工具中内容结构的建模或表示方式并不相同。此外，工具还使用不同的底层存储机制和格式。
As noted, not all PKG tools take the same approach. This means that content structure is not modeled or represented in the same way across tools. Furthermore, tools also use different underlying storage mechanisms and formats.
由于 PKG 是一个新兴领域，因此没有标准。事实上的唯一标准是 Markdown，因为它可以作为 PKG 工具中数据导入的中间格式，而 PKG 工具本身不支持存储格式。然而，目前，即使是 Markdown 也是一种最不常见的格式。
As PKG is a nascent domain, there are no standards. The only de facto standard is Markdown, as it can serve as an intermediate format for data import in PKG tools where not supported natively as the storage format. At present, however, even Markdown is a least common denominator format.
数据位置
Data location
对于存储，有一些重要的考虑因素：谁拥有和控制数据？数据存储在哪里？如何存储？
For storage there are some important considerations: Who owns and controls the data? Where is it stored? How is it stored?
根据数据的存储方式，我们可以将数据位置大致分为三类：
We can broadly separate the data location into three categories depending on how data is stored:
1. 用户硬盘上的非专有文件
1. Nonproprietary files on the user’s hard drive
2. 用户硬盘上的专有文件或数据库
2. Proprietary files or a database on the user’s hard drive
3. 云服务中的数据库
3. A database in a cloud service
文件通常会导致用户拥有数据所有权，而在线云数据库可以实现同步和协作功能。
Files most often result in data ownership for the user, while online cloud databases enable synchronization and collaboration features to be implemented.
互操作性
Interoperability
在缺乏标准的情况下，从 Markdown 导出和导入数据的能力在很大程度上决定了每种工具的互操作能力。此外，我们还会考虑导入和导出其他格式的数据的能力，例如 CSV、HTML 和 PDF。无论如何，PKG 工具的互操作性都有很多不足之处。
In the absence of standards, the ability to export and import data to and from Markdown largely defines the interoperability capabilities of each tool. Additionally, we consider the ability to import and export data in other formats, such as CSV, HTML and PDF. Either way, the interoperability of PKG tools leaves a lot to be desired.
不同的工具以不同的方式建模数据，这是分歧点。即使是相同的数据模型也可以用不同的 Markdown 语法变体来表示。此外，不同工具在数据结构上也存在功能差异。因此，每种工具都带有自己的 Markdown 风格以适应其数据模型，而有些工具根本不支持 Markdown。
The fact that different tools model their data in different ways is a divergence point. Even the same data model can be represented in different Markdown syntax variations. Furthermore, there are also differences in functionality reflected in data structure among tools. As such, each tool comes with its own flavor of Markdown to accommodate its data model, while some tools don’t support Markdown at all.
情态
Modality
过去几年，云已成为应用程序开发和数据存储的最常见选择。PKG 工具也不例外：它们中的大多数都在云中运行和存储数据。有些提供云和本地选项，而只有少数不提供云版本。大多数 PKG 工具还提供移动应用程序，可随时随地使用它们。
Over the last few years, the cloud has become the most common option for both application development and data storage. PKG tools are no exception: most of them operate and store their data in the cloud. Some offer both cloud and on-premises options, while only a minority do not offer a cloud version. Most PKG tools also offer mobile apps that enable using them on the go.
虽然有些用户可能更喜欢云软件提供的易用性，但也有几个原因导致其他人更喜欢本地软件。数据所有权和控制权是一个原因。定价是另一个原因，因为基于云的服务几乎肯定意味着基于订阅的模式。
While some users may prefer the ease of use that cloud software offers, there are also several reasons why others prefer on-premises software. Data ownership and control is one reason. Pricing is another, as a cloud-based service almost certainly means a subscription-based model.
同步
Synchronization
对于使用提供多种不同模式的工具的用户来说，在不同平台之间同步数据的能力至关重要。它确保无论在哪个平台上使用该工具，都可以获得最新更新。
The ability to synchronize data among different platforms is key for users of tools that offer many different modalities. It ensures that the latest updates will be available regardless of what platform the tool is used on.
PKG 工具提供基于云的服务的一个优势是，这使得同步更加容易。同步通常作为一项增值服务提供。或者，也可以使用量身定制的方法和外部基于云的服务（例如 Google Drive、Microsoft OneDrive 或 GitHub）来实现同步。然而，对于普通用户来说，这并不总是容易实现的，它可能很笨重且容易出错，并且不能解决数据所有权问题。
An advantage for PKG tools offering a cloud-based service is that this makes synchronization easier. Synchronization is often offered as a value-add service. Alternatively, synchronization can also be implemented using tailor-made approaches and external cloud-based services, such as Google Drive, Microsoft OneDrive or GitHub. However, this is not always easy to implement for average users, it can be clunky and error prone, and does not address the data ownership issue.
价格
Pricing
对于大多数用户来说，这是一个重要的考虑因素。尽管他们活跃在同一市场，但 PKG 工具供应商在多个方面都采取了不同的做法。他们在技术和其他方面做出了各种不同的选择。他们利用不同的感知优势和叙述来满足不同的用户群体。
This is an important consideration for most users. Although they are active in the same market, PKG tool vendors approach it differently in several ways. They make various divergent choices, technical and otherwise. They address different user segments using different perceived advantages and narratives.
因此，供应商的定价政策也各不相同。由于几乎所有供应商都提供其工具的云版本，因此他们还提供具有不同层级的订阅模式。大多数工具还提供免费层级。但是，免费层级的内容和作为单独付费服务提供的内容有所不同。
As a result, vendor pricing policies also vary. As almost all vendors offer cloud versions of their tools, they also offer subscription models with different tiers. Most tools also offer a free tier. However, what is part of the free tier and what is offered as a separate paid service varies.
执照
License
对于许多用户来说，价格并不是最重要的。虽然大多数 PKG 工具都提供免费套餐，但那些看重透明度、所有权和社区价值的用户会选择开源工具。开源不仅意味着免费使用和拥有，还意味着免费修改和审核。开源可以促进软件开发以及社区和用户群的增长，但可以说，这让建立可持续业务变得更加困难。
For many users, pricing is not all that matters. While most PKG tools offer a free tier, users who rank transparency, ownership and community values high will want to opt for open source tools. Open source does not only mean free to use and own, but also free to modify and audit. Open source can facilitate software development as well as community and user base growth, but arguably makes it harder to build a sustainable business.
开源 PKG 工具属于少数。它们采用类似的盈利策略。它们主要提供附加组件或服务，例如收费同步。它们还利用支持者的自愿贡献来换取福利，例如访问早期版本或社区徽章。
Open source PKG tools are in the minority. They adopt similar monetization strategies to each other. Primarily, they offer add-ons or services, such as synchronization for a fee. They also leverage voluntary supporter contributions in exchange for perks, such as access to early builds or community badges.
生态系统
Ecosystem
工具的生态系统是其整体效用的一个重要参数。生态系统是指工具可用的外围功能和支持的集合。这归结为两点：每个工具周围的社区有多大和活跃，以及工具的架构和供应商的政策是否促进了额外模块的开发，从而开发出有用的插件。
A tool’s ecosystem is an important parameter in its overall utility. By ecosystem we mean the collection of peripheral functionality and support available for a tool. This comes down to two things: how big and active the community is around each tool, and whether the tool’s architecture and the vendor’s policies facilitate the development of additional modules, resulting in the development of useful plugins.
这两个参数是相互独立的。通常，培育生态系统被视为开源的必然结果。然而，情况并非如此。有些开源解决方案没有很好的生态系统，而有些专有解决方案却有。第三个参数是存在 API，以促进与其他应用程序和数据源的集成。
Those two parameters are orthogonal. Often, nurturing an ecosystem is seen as something that comes with being open source. However, that is not necessarily the case. There are examples of open source solutions that do not have a great ecosystem, as well as proprietary solutions that do. A third parameter is the existence of an API to facilitate integration with other applications and data sources.
供应商信誉
Vendor credibility
在评估任何解决方案时，供应商的可信度始终是一个重要参数。我的工具是否会不断发展并享受定期更新？如果我有问题或请求，它会被考虑，有人会回复我吗？我的工具会长期存在吗？我能指望它的制造商履行他们的承诺吗？所有关键问题的答案都归结为供应商的可信度。
Vendor credibility is always an important parameter when evaluating any solution. Will my tool evolve and enjoy regular updates? If I have an issue or request, will it be considered, and will someone get back to me? Will my tool be around for the long run? Can I expect its makers to honor their commitments? All key questions, the answers to which come down to vendor credibility.
这是一个新兴领域，工具之间的切换并不容易。因此，供应商锁定是需要注意的问题，用户应该明智地选择供应商。我们根据几个参数来评估供应商的可信度，例如供应商存在的时间、资金、员工人数、商业模式和吸引力。
This is a nascent domain and switching between tools is not easy. Consequently, vendor lock-in is something to be aware of and users should choose a vendor wisely. We evaluate vendor credibility based on several parameters, such as how long they have been around, funding, headcount, business model and traction.
功能参数
Functional parameters
对于一些 PKG 工具制造商和用户来说，他们的目标似乎是拥有一个“万能平台”。也就是说，它不仅可以做笔记，还可以做从日历和任务管理到电子邮件和社交媒体集成等所有事情。
For some PKG tool makers and users, the goal seems to be to have an “everything platform.” That is, not just to take notes, but to be able to do everything from calendar and task management to email and social-media integration.
这符合操作系统的趋势，即日历、邮件、联系人、笔记等多个个人信息管理应用程序相互关联，可通过一个智能搜索进行访问。为了实现这一点，PKG 工具需要非常灵活且可通过其生态系统进行扩展，或者提供多种功能。
This follows a trend in operating systems, where several applications for personal information management, such as calendar, mail, contacts, notes, etc., are interlinked and can be accessed from one smart search. To enable this, PKG tools need to either be very flexible and extendable through their ecosystem, or offer a multitude of features.
以下是 PKG 工具提供的功能列表。
The following is a list of features offered by PKG tools.
文本编辑
Text editing
所有这些工具都提供了 Markdown 编辑器所具备的一般文本编辑功能。具有专有数据存储的工具有时会提供多列布局、突出显示和文本颜色等高级功能。有趣的是，这些工具还缺少文本编辑器的一些基本功能，例如搜索和替换。
All these tools offer general text editing capabilities expected of Markdown editors. Tools with proprietary data storage sometimes offer advanced features like multicolumn layouts, highlighting and text colors. Interestingly, these tools also lack some fundamental features of text editors like search and replace.
实时协作
Real-time collaboration
实时协作是指不同用户以类似 Google Docs 的方式在 PKG 工具中编辑内容的能力。要实现这一点，需要支持多个用户。此功能还依赖于版本控制等功能。目前，大多数工具的协作功能都与人们熟悉的 Google Docs “基线”相差甚远。
Real-time collaboration is the ability for different users to edit content in PKG tools in a way similar to how Google Docs works. For this to be feasible, multiple users need to be supported. This feature also relies on capabilities, such as versioning. At present, the collaborative features of most tools are not close to the Google Docs “baseline” that people are familiar with.
图形视图
Graph view
图形视图允许用户通过单击图形中的实体或关系来直观地导航 PKG。单击可在编辑器中打开它们。这对于导航和概览 PKG 的链接数和内容非常有用。图形视图因 PKG 增加节点和连接数量而变得不那么有用而受到批评。
A graph view allows the user to navigate their PKG visually by clicking on entities or relationships in their graph. Clicking opens them in the editor. This is useful for navigating and getting an overview of the link count and content of the PKG. Graph views have been criticized for becoming less useful as PKGs increase the number of nodes and connections.
日记
Journaling
许多 PKG 工具都提供了每日记笔记的界面，而不是从空白笔记开始。有证据表明日记有积极作用，但这种方法还有将笔记与日期联系起来的好处，即使在没有参考资料的笔记上也能建立图表联系。
Instead of starting on a blank note, many PKG tools offer an interface for daily note-taking. There is evidence of positive effects of journaling, but this approach also has the benefit of linking notes to dates, getting the graph connections going even on notes that have no references.
任务规划
Task planning
不同工具的任务规划功能各不相同。有些工具仅允许在笔记上设置提醒，而有些工具则允许进行复杂的任务管理，包括任务分配、截止日期、协作等。虽然大多数工具都支持个人任务管理，但它们的实时协作通常还不够成熟。
Task planning capabilities vary between tools. While some of them only allow setting reminders on notes, others allow complex task management with assignments, deadlines, collaboration, etc. While personal task management is possible in most tools, their real-time collaboration is typically not mature.
结构化数据
Structured data
一些 PKG 工具提供数据库功能，这意味着它们可以存储和查询结构化数据。这是一项高级功能，非常强大且灵活。特别重要的是能够支持图形查询语言，从而可以访问图形模式和路径。
Some PKG tools offer database functionalities, meaning they can store and query structured data. This is an advanced feature, and very powerful and flexible. Of particular relevance is the ability to support a graph query language, enabling access to graph patterns and paths.
出版
Publishing
我们将发布定义为在网络上共享使用 PKG 工具创建的内容的能力。根据其方法和模式，一些工具本身提供此功能，而其他工具则将其作为附加服务提供。
We refer to publishing as the ability to share content created in PKG tools on the web. Depending on their approach and modality, some tools offer this capability natively while others offer it as an additional service.
推介会
Presentation
有些工具有原生的显示模式。用户可以对显示的笔记或区块进行排序，而不是使用固定的页面，还可以进行导航。
Some tools have native presentation modes. Instead of fixed pages, users can order the notes or blocks that are shown, and they can also be navigated.
间隔重复
Spaced repetition
间隔重复是一种基于证据的学习技术，通常使用抽认卡进行。[3] 新引入的较难的抽认卡显示频率较高，而较旧且较不难的抽认卡显示频率较低，以利用心理间隔效应。研究表明，使用间隔重复可以提高学习速度（Smolen 等人，2016 年）。
Spaced repetition is an evidence-based learning technique that is usually performed with flashcards. [3] Newly introduced and more difficult flashcards are shown more frequently, while older and less difficult flashcards are shown less frequently to exploit the psychological spacing effect. Research suggests that the use of spaced repetition increases the rate of learning (Smolen et al., 2016).
间隔重复是 RemNote 的重点。Logseq 还支持抽认卡。虽然其他工具可以通过提醒或插件实现间隔重复，但只有 RemNote 支持创建带有内置计时器和用于提问的用户界面的专用抽认卡。
Spaced repetition is a focus of RemNote. Logseq also provides support for flashcards. While spaced repetition can be achieved in other tools through reminders or plugins, only RemNote supports the creation of dedicated flashcards with inherent timers and a user interface for questioning.
PDF 和其他外部文件
PDF and other external files
PDF 无处不在，尤其是在研究领域。由于许多 PKG 工具面向知识工作者，因此大多数工具都允许某种形式的 PDF 导入或注释。然而，此功能凸显了这些工具中专有数据格式的集中方法有多大的问题。
PDFs are ubiquitous, especially in research. As many of these PKG tools are marketed towards knowledge workers, most of them allow some kind of PDF import or annotation. This feature however highlights how problematic the centralized approach of proprietary data formats in these tools has become.
在撰写本文时，没有工具允许在文件系统上编辑和突出显示 PDF，而无需在每次进行更改时导入和导出文件。这意味着在该工具中编辑的 PDF 将无法与任何其他 PDF 软件共享或互操作。此限制也适用于可以导入的许多其他文件格式。
At the time of writing no tool allows editing and highlighting PDFs on the file system without having to import and export the file every time changes are made. This means PDFs edited in the tool will not be shareable or interoperable with any other PDF software. This limitation also applies to many other file formats that can be imported.
工具评估
Tool evaluation
许多工具都可以归类为笔记工具，它们可以创建类似于 PKG 的结构。我们根据受欢迎程度和相关性选择了以下内容，同时还涵盖了广泛的卖点和目标受众。
Many tools could be categorized as note-taking tools that enable the creation of structures akin to PKGs. We have selected the following based on popularity and relevance, while also covering a broad spectrum of selling points and target audiences.
项目符号大纲
Bullet Outliners
雷姆诺特
RemNote
RemNote 是一家专注于学术和记忆工作的大纲编写器。它成立于 2020 年。
RemNote is an outliner with a focus on academic and memory work. It was founded in 2020.
该公司于 2021 年获得 300 万美元的种子轮融资，目前拥有 16 人的团队。[4] RemNote 拥有 100,000+ 注册用户和约 15,000+ 每日活跃用户 (DAU)。
The company has received a seed funding round of $3 million in 2021 and has a team of 16 people. [4] RemNote has 100,000+ registered users and approximately 15,000+ Daily Active Users (DAU).
RemNote 是一款专有软件。它可以通过浏览器或适用于 Windows、macOS、Linux、Android 和 iOS 的桌面和移动应用程序使用。RemNote 支持 Markdown、OPML、JSON、HTML 和 Anki 格式的导入和导出以及纯文本导出，并提供 API。
RemNote is proprietary software. It can be used via browsers or desktop and mobile applications available for Windows, macOS, Linux, Android, and iOS. RemNote supports import and export in Markdown, OPML, JSON, HTML and Anki formats as well as plain text export and it offers an API.
RemNote 提供了一些独特的功能，即高级 pdf 注释和间隔重复抽认卡系统。
RemNote offers some unique features, namely advanced pdf annotation and the spaced repetition flashcard system.
工作流程
Workflowy
Workflowy 是一家专注于“简洁和优雅”的大纲编写公司。该公司成立于 2010 年。
Workflowy is an outliner with a focus on “simplicity and elegance.” It was founded in 2010.
该公司在 2010 年和 2017 年获得了两轮种子融资，金额不详，目前员工人数不多，截至撰写本文时为 12 人。Workflowy 的免费和付费计划共有 300 万用户。
The company has received two seed funding rounds of undisclosed amounts in 2010 and 2017 and it employs a handful of people – 12 at the time of writing. Workflowy boasts a total of 3 million users combined in its free and premium plans.
Workflowy 是专有软件。它可以通过浏览器或适用于 Windows、macOS、Linux、Android 和 iOS 的桌面和移动应用程序使用。Workflowy 支持以纯文本和格式化文本、OPML 和 Guide JSON 格式导出。Markdown 导出本身不受支持，只能通过变通方法或第三方扩展来支持。完全不支持 Markdown 导入。
Workflowy is proprietary software. It can be used via browsers or desktop and mobile applications available for Windows, macOS, Linux, Android, and iOS. Workflowy supports export in plain and formatted text, OPML and Guide JSON formats. Markdown export is not supported natively, only via workarounds or a third-party extension. Markdown import is not supported at all.
Workflowy 已经存在很长一段时间了，并且启发了此列表中的许多其他工具，这就是它被列入的原因。
Workflowy has been around for quite a while and has inspired many of the other tools in this list, which is why it is included.
塔那
Tana
Tana 的愿景是“重塑人类、团队和计算机的协同工作方式”。该公司成立于 2020 年。
Tana’s vision is to “reinvent how humans, teams and computers work together.” The company was founded in 2020.
Tana 于 2020 年从多位天使投资者和风险投资公司获得了一轮未披露金额的种子资金。该公司拥有 17 名员工。在撰写本文时，Tana 仍处于早期阶段。据说它将支持数据导出以及免费计划。
Tana received a seed funding round of an undisclosed amount in 2020 from a number of angel investors and VCs. The company employs 17 people. At the time of writing, Tana is still in early access. It is said that it will support data export as well as a free plan.
Tana 带来了新颖的功能和独特的用户界面。Tana 将基于对象/模式的方法（如 Notion 或 Airtable）与功能齐全的大纲相结合。它是唯一一款在关系中添加逻辑和语义（包括传递性）的工具。
Tana brings novel features and a distinctive UI to the table. Tana combines an object/schema-based approach, like Notion or Airtable, with a fully functional outliner. It is the only tool adding logic and semantics in the relationships, including transitivity.
Markdown 编辑器
Markdown Editors
黑曜石
Obsidian
Obsidian 是一款可扩展的 Markdown 编辑器，专注于数据所有权和控制权。它于 2020 年启动。
Obsidian is an extensible Markdown editor with a focus on data ownership and control. It was initiated in 2020.
目前没有关于 Obsidian 背后实体的财务状况或员工人数的公开信息。它的两位创始人也是其发展的推动力，他们之前曾使用过大纲工具 Dynalist。最近招募了一位新的首席执行官，团队正在扩大。[5] Obsidian 是一款开源软件，拥有一个围绕其插件生态系统建立的充满活力的贡献者社区。在撰写本文时，Obsidian 的 Discord 拥有 70,000 名成员。
There is no public information available on the financials or headcount of the entity behind Obsidian. Its two founders are also the driving force behind its development, with previous experience with the outliner Dynalist. A new CEO was recently recruited and the team is expanding. [5] Obsidian is open source software with a vibrant community of contributors built around its plugin ecosystem. At the time of writing, Obsidian’s Discord has 70,000 members.
Obsidian 提供适用于 Windows、Linux 和 macOS 的桌面客户端，以及适用于 Android 和 iOS 的移动应用程序。该软件没有基于 Web 的版本。
Obsidian provides desktop clients for Windows, Linux, and macOS, as well as mobile apps for Android and iOS. There is no web-based version of the software.
Obsidian 将其内容存储在 Markdown 中，因此原生支持 Markdown 导入和导出。唯一的例外是其 Canvas 功能和多媒体文件。其他格式也通过插件支持。Obsidian 的插件生态系统和 API 是其日益流行的关键部分。可用的插件可以适应越来越多的用例，例如项目管理、日历等。
Obsidian stores its content in Markdown, so Markdown import and export are supported natively. The only exception is its Canvas feature and multimedia files. Other formats are also supported via plugins. Obsidian’s plugin ecosystem and API are a key part of its rising popularity. Available plugins can accommodate a growing list of use cases, such as project management, calendar, and more.
Obsidian 可免费供个人使用。我们鼓励用户捐款或购买受支持的许可证和/或访问 Obsidian 的数据同步服务以及内置的互联网发布服务。商业许可证每位用户每年收费 50 美元，并包括优先客户支持。
Obsidian is free for personal use. Users are encouraged to make donations or purchase a supported license and/or access to Obsidian’s data synchronization service, as well as a built-in internet publishing service. The commercial license costs $50 per user, per year, and includes prioritized customer support.
对数序列
Logseq
Logseq 是一个开源 Markdown 大纲编辑器，其“受到 Org Mode、Roam、TiddlyWiki 和 Workflowy 的强烈启发”。它强调数据所有权，将数据存储在纯 Markdown 文本文件中。它成立于 2021 年。
Logseq is an open source Markdown outliner that is “strongly inspired by Org Mode, Roam, TiddlyWiki and Workflowy.” It places an emphasis on data ownership, storing its data in plain Markdown text files. It was founded in 2021.
2022 年，该公司从多家投资者那里获得了 410 万美元的种子轮融资，其中包括著名的风险投资公司 Andreessen Horowitz。虽然没有关于其用户数量和员工人数的信息，但其 GitHub 存储库拥有近 20,000 颗星。Logseq 目前不提供付费计划，但可以肯定的是，它最终会提供付费计划。鼓励用户捐款。
In 2022, the company received a $4.1 million seed funding round from a number of investors, including the notable VC firm Andreessen Horowitz. While information about its numbers of users and headcount is not available, its GitHub repository boasts nearly 20,000 stars. Logseq does not offer paid-for plans at this time, but it is safe to assume it will, eventually. Users are encouraged to make donations.
Logseq 提供适用于 Windows、Linux 和 macOS 的桌面客户端，以及适用于 Android 和 iOS 的移动应用程序。该软件没有基于 Web 的版本。Logseq 提供了一个 API，人们用它来开发插件。它通过插件（仍在开发中）支持导出为 PDF 和其他格式的功能，
Logseq provides desktop clients for Windows, Linux, and macOS, as well as mobile apps for Android and iOS. There is no web-based version of the software. Logseq offers an API that people are using to develop plugins. It is through plugins (still in development) that it supports export functionality to PDF and other formats,
Logseq 正在利用其开源方向和当前的吸引力来构建其生态系统，目前包括 100 多个插件和 30 多个主题。Logseq 使用户能够创建任务、管理和存储笔记或待办事项列表、嵌入页面和注释 PDF。其既定目标是开源知识协作并创建世界知识图谱。
Logseq is leveraging its open source direction and current traction to build its ecosystem, which currently includes 100+ plugins and 30+ themes. Logseq enables users to create tasks, manage and store notes or to-do lists, embed pages and annotate PDFs. Its stated goal is to open source knowledge collaboration and create a world knowledge graph.
块接口
Block Interfaces
概念
Notion
Notion 是一个基于块的编辑器，自 2013 年以来一直存在。这是 Notion Labs, Inc. 成立的年份。
Notion is a block-based editor that has been around since 2013. This is the year Notion Labs, Inc. was founded.
自那时以来，该公司在五轮融资中筹集了总计 3.432 亿美元，投资者包括 Index Ventures 和 Sequoia Capital。Notion 还进行了四次收购，这有助于扩大其功能。2021 年，Notion 拥有超过 2000 万用户，估值为 100 亿美元。该公司拥有数百名员工。
Since then, the company has raised a total of $343.2 million in five funding rounds involving VC investors, such as Index Ventures and Sequoia Capital. Notion has also made four acquisitions, which have helped grow its functionality. In 2021, Notion had over 20 million users and a valuation of $10 billion. The company employs a few hundred people.
Notion 可通过浏览器或桌面和移动应用程序使用，适用于 Windows、macOS、Linux、Android 和 iOS。它有四层订阅模式：免费、个人、团队和企业。Notion 是专有软件。它的导出功能包括 PDF、HTML、CSV 和 Markdown。还可以导入 Markdown。
Notion can be used via browsers or desktop and mobile applications available for Windows, macOS, Linux, Android, and iOS. It has a four-tiered subscription model: Free, Personal, Team and Enterprise. Notion is proprietary software. Its export capabilities include PDF, HTML, CSV and Markdown. Markdown can also be imported.
Notion 提供了广泛的开箱即用功能以及 API 和与许多应用程序的集成。它的用例包括个人知识管理、项目和团队管理、实时协作和内容发布。但是，它的图形特定功能很少。此外，在 Notion 中，与 Roam、Logseq 和 Tana 不同，块不是节点，这意味着它们始终属于层次结构中的页面，不能作为页面或实体打开。
Notion offers a wide range of functionality out of the box as well as an API and integrations with many applications. Its use cases include personal knowledge management, project and team management, real-time collaboration and content publishing. However, its graph-specific features are minimal. Also in Notion, unlike Roam, Logseq, and Tana, the blocks are not nodes, meaning they always belong to a page in a hierarchy and cannot be opened as pages or entities.
漫游研究
Roam Research
Roam Research 是一家专注于研究的专有区块大纲制定者。该公司成立于 2017 年。
Roam Research is a proprietary block outliner with a focus on research. It was founded in 2017.
2020 年，该公司从多位天使投资者和风险投资公司获得了 900 万美元的种子轮融资。该公司拥有“11 至 50 名”员工。2020 年，Roam 估计拥有 60,000 名用户，估值为 2 亿美元。[6]
The company received a seed funding round of $9 million from a number of angel investors and VCs in 2020. The company employs “between 11 and 50” people. In 2020, Roam was estimated to have 60,000 users and a $200 million valuation. [6]
Roam 是一款专有软件。它提供适用于 Windows、Linux 和 macOS 的桌面客户端，以及适用于 Android 和 iOS 的移动应用程序。该软件还有一个基于 Web 的版本。其导出功能包括 Markdown、JSON 和 EDN，这是一种无损图形备份格式。导出和导入都有一些限制。API 正在规划中。
Roam is proprietary software. It provides desktop clients for Windows, Linux, and macOS, as well as mobile apps for Android and iOS. There is also a web-based version of the software. Its export capabilities include Markdown, JSON and EDN, which is a format for lossless graph backups. Both export and import have some limitations. An API is on the roadmap.
Roam 推广并推销了所谓的“思维工具”类别，并引入了个人知识图谱的许多方面。它拥有一批忠实的追随者，有时也被称为“邪教”，他们构建工作流、主题和插件。Roam 只提供付费订阅。它启发了 Logseq 和 Athens 等免费开源替代方案。
Roam popularized and marketed the so-called “tools for thought” category and introduced many aspects of personal knowledge graphs. It has a devoted following, sometimes also referred to as a “cult,” that builds workflows, themes and plugins. Roam only offers a paid subscription. It has inspired free and open source alternatives like Logseq and Athens.
比较
Comparison
在下表中，我们展示了每种工具在我们评估的功能中的表现。
In the following tables, we present how each tool fares across the features we evaluate.
非功能参数比较
Nonfunctional parameter comparison
表 3.1 非功能参数比较
Table 3.1 Nonfunctional parameter comparison
功能参数对比
Functional parameter comparison
表3.2 功能参数对比
Table 3.2 Functional parameter comparison
评估
Evaluation
我们刻意避免对“最佳”工具做出陈述。此类陈述具有主观性，需要结合上下文信息进行评估。不过，我们会根据对 PKG 工具不同方面的评估提出建议。
We deliberately refrain from making statements about what the “best” tool is. Such statements are subjective and need to be evaluated with contextual information in mind. However, we present recommendations based on evaluating different aspects of PKG tools.
隐私和数据所有权
Privacy and data ownership
Obsidian 和 Logseq 使用本地存储的基本 Markdown 文本文件，可以使用任何文件浏览器进行浏览。为了获得高级功能，它们在这些文本文件上运行算法，以构建具有双向链接的图表。这种方法的优点是用户拥有完全的数据控制权和所有权，以及与所有其他 Markdown 或文本编辑器的互操作性——至少在基本层面上。
Obsidian and Logseq use basic Markdown text files that are stored locally and can be browsed with any file explorer. To get their advanced functionality, they run algorithms on these text files to build a graph with the bidirectional links. This approach has the advantage of complete data control and ownership for the user, as well as interoperability with all other Markdown or text editors – at least on a basic level.
其他工具则将数据存储在专有云服务中。这意味着用户没有数据控制权和所有权，为了以可互操作的格式获取数据，他们必须将其导出。然而，这确实使这些工具能够轻松实现自己的跨设备同步。
The other tools store their data proprietarily and in cloud services. This means the user does not have data control and ownership, and to get their data in interoperable formats they have to export it. It does, however, enable these tools to easily implement their own cross-device synchronization.
功能和可扩展性
Functionality and extensibility
这方面很大程度上取决于每种工具的理念。大纲编辑器提供以笔记记录为中心的最低限度的功能。Markdown 编辑器和块界面都提供广泛的功能，尽管它们各自以自己的方式实现。
This aspect is very much dependent on the philosophy of each tool. Outliners offer minimal functionality centered around note-taking. Markdown editors and block interfaces all offer extensive functionality, although each does it in its own way.
Notion 是一个具有许多功能的工具的例子，而 Roam Research 和 Obsidian 则是提供大量定制和可扩展性的工具。
Notion is an example of a tool that has many features, while Roam Research and Obsidian are tools that offer a lot of customization and extensibility.
Roam Research、Obsidian 和 Logseq 提供了很多自定义功能，既包括设置和数据存储机制，也包括通过公开可用的扩展和插件。2022 年，Roam Research 采用了一种新的商业模式，公司根据扩展的使用量向扩展开发人员支付报酬。这既影响了（部分）扩展的控制，也影响了其质量。
Roam Research, Obsidian and Logseq offer a lot of customization, both in settings and data storage mechanisms, as well as through openly available extensions and plugins. In 2022, Roam Research applied a new business model in which extension developers are paid by the company depending on how much their extensions are used. This influenced both the control and the quality of (some of) the extensions.
用户体验和整体完善
User experience and overall polish
对于想要成熟精致的用户界面的人来说，Notion 是一个不错的选择。它以出色的用户界面和功能覆盖范围脱颖而出，并提供良好的结构化数据、发布和协作功能。
For people that want a mature and polished user interface, Notion is a safe recommendation. It stands out with great user interface and feature coverage and offers good structured data, publish and collaboration functionality.
对于一款快速而简约的大纲编写器来说，Workflowy 非常精致，但提供的功能集较少。
For a fast and minimalistic outliner, Workflowy is very polished but offers a reduced feature set.
发布和实时协作
Publishing and real-time collaboration
Notion 和 Logseq 提供免费发布，而 Roam 和 Obsidian 则提供付费发布服务。
Notion and Logseq offer free publishing, while Roam and Obsidian offer it as a paid service.
Notion、Roam 和 Workflowy 也支持实时协作。大多数其他工具在其路线图或测试版中都包含协作功能。
Notion, Roam and Workflowy also support real-time collaboration. Most of the other tools do have collaboration in their roadmap or beta.
高级图表功能
Advanced graph features
就实际的图表部分而言，其突出程度在不同工具之间差异很大。在大纲中，它是后台的一个抽象，不会向用户展示太多。虽然它们的数据模型可能与其他笔记工具不同，但实际上，除了对反向链接的支持外，传统笔记工具与 PKG 大纲几乎没有什么区别。尽管 Notion 功能丰富，但情况也是如此。
As far as the actual graph part goes, its prominence varies wildly across tools. In outliners, it is an abstraction in the background which does not get much exposure to users. Although their data model may differ from other note-taking tools, in practice, there is little separating legacy note-taking tools from PKG outliners besides support for backlinks. Similar things could be said for Notion, despite its rich functionality.
Roam、Obsidian、Logseq 和 Tana 将图表作为其产品的核心部分。它们都提供图表导航或图表视图，以及高级图表查询功能。
Roam, Obsidian, Logseq and Tana make graphs a central part of their offering. They all offer graph navigation or graph views, as well as advanced graph querying capabilities.
定价和访问
Pricing and access
这可能是所有主观比较中最主观的，因为感知的价格与价值比取决于许多因素。工具可能价格相似，但属于不同的类别或提供不同的功能集。此外，用户对功能的评价也不同。
This is perhaps the most subjective among all subjective comparisons, as the perceived price to value ratio depends on a number of factors. Tools may be similarly priced but belong to different categories or offer a different set of features. In addition, users value features differently.
大多数 PKG 工具都采用类似的方法。它们采用具有多个层级的订阅模式，除 Roam 外，所有工具都提供功能有限的免费层级。不过，Roam 还提供协作图表，只有所有者才需要许可证。由于 Tana 处于早期访问阶段，因此定价和平台尚未确定。
Most PKG tools have a similar approach. They use a subscription model with several tiers, with all except Roam offering a free tier with limited functionality. However, Roam also offers collaborative graphs, for which only the owner needs a license. As Tana is in early access, the pricing and platforms are not fixed yet.
目前，唯一部分免费的工具是 Obsidian 和 Logseq。Obsidian 只对同步和发布等附加服务收费。这些服务可以绕过，可能不是所有用户都感兴趣，但它们的价格并不便宜。此外，将 Obsidian 用于商业用途需要年度许可。Obsidian 也像其他工具一样在存储方面没有任何限制，因为它不会在云中存储任何东西。Logseq 和 Tana 也必然会对附加服务收费。
Currently, the only tools that are partially free to use are Obsidian and Logseq. Obsidian only charges for add-on services like synchronization and publishing. Those can be worked around and may not be of interest to all users, however their pricing is not cheap. Also, a yearly license is required to use Obsidian for commercial purposes. Obsidian also does not have any limitations in terms of storage like other tools because it does not store anything in the cloud. Logseq and Tana are bound to introduce charges for add-on services too.
结论与展望
Conclusion and outlook
个人知识图谱的概念是一个新兴概念，这反映在相应的工具类别中。在所有软件类别中，边界都有些模糊；这在新兴领域更为明显。
The concept of personal knowledge graphs is an emergent one and this is reflected in the corresponding category of tools. In all software categories the boundaries are somewhat fluid; this is even more pronounced in emergent fields.
目前，PKG 工具类别并不单一。它汇集了具有不同起源、理念和目标受众的工具，并通过一套松散的共同原则和基本原理将它们结合在一起：提升连接、创建双向链接以及以图形形式导航和可视化笔记。
At present, the PKG tool category is not homogeneous. It aggregates tools with very different origins, philosophies and target audiences, with a loose set of common principles and primitives bringing them together: elevating connections, creating bidirectional links, and navigating and visualizing notes as a graph.
企业知识图谱和个人知识图谱的出现存在一些有趣的相似之处。探索它们可以为 PKG 如何发展提供一些见解。
There are some interesting parallels in the emergence of enterprise knowledge graphs and personal knowledge graphs. Exploring them can provide some insights as to how PKGs may evolve.
对于企业而言，大量不同的应用程序（每个应用程序都有自己的数据库和数据模型）的激增导致了碎片化。整合这种碎片化格局需要持续和临时地付出大量努力。对于企业而言，其动机来自于对应用程序互操作性和分析的整体视图的需求。
In the case of enterprises, the proliferation of many different applications, each with its own database and data model, leads to fragmentation. Integrating this fragmented landscape requires a lot of effort in a continuous and ad hoc way. For enterprises the motivation comes from wanting application interoperability and a holistic view for analytics.
知识图谱非常适合数据集成，这促使许多企业采用它们。采用知识图谱范式还意味着在数据管理方面采用更具原则性的方法，包括元数据、词汇和概念管理。企业知识图谱平台可帮助企业加入并在这些方面为其提供支持。
Knowledge graphs are ideally suited for data integration, and this has led many enterprises to adopt them. Adopting the knowledge graph paradigm also means adopting a more principled approach around data management, including metadata, vocabulary and concept curation. Enterprise knowledge graph platforms help enterprises get on board and support them in these aspects.
在个人使用的情况下，应用程序蔓延的问题较少。对于大多数用户来说，知识图谱的入口点似乎是通过笔记记录而隐蔽的。PKG 工具的受众目前似乎主要包括知识工作者，这并非巧合：这些人要么在工作中大量参与笔记记录（研究人员和创意人员），要么是笔记记录和生产力爱好者。
In the case of personal use, application sprawl is less of a problem. For most users the entry point to knowledge graphs seems to be a covert one via note-taking. It’s no coincidence that the audience for PKG tools at this point seems to mostly include knowledge workers: people who either engage in note-taking extensively as part of their work (researchers and creatives) or are note-taking and productivity enthusiasts.
后一部分受众也是推动“一切皆平台”观点的人。对于企业而言，采用以数据为中心的方法以及应用程序开发和集成的最佳实践可以带来许多切实的好处。对于个人使用而言，进行这样的尝试超出了大多数用户的能力范围，并且可能不会产生同等的好处。
This latter part of the audience is also the one driving the “everything-platform” view. For enterprises, adopting a data-centric approach and best practices for application development and integration can have many tangible benefits. For personal use, embarking on such an endeavor is something beyond the capacity of most users and would probably not yield comparable benefits.
虽然有些用户接受“万物平台”的观点，但即使是简单的笔记也缺乏标准和互操作性，这意味着采用一个平台来处理所有事情可能会导致供应商锁定。
While some users embrace the “everything-platform” view, the lack of standards and interoperability even for simple notes means that adopting one platform for everything would probably result in vendor lock-in.
在新兴领域，比如这个领域，对仍在寻找产品市场契合度并争夺市场份额的供应商进行过多投资可能并不明智。我们已经有一个供应商（Athens Research）的例子，该供应商已停止开发其核心 PKG 工具，并在获得种子资金并获得一些关注后不久转向转型。
In a nascent domain, such as this one, investing too heavily in vendors still looking for product-market fit and vying for market share would probably be unwise. We already have an example of a vendor (Athens Research) that has stopped developing its core PKG tool and is heading for a pivot shortly after receiving a seed funding round and gaining some traction.
开放和企业知识图谱的一个重要方面是语义支持，而目前 PKG 工具几乎完全没有这种支持。具有正式定义的受控词汇表和具有明确语义的类型化关系/链接是开放知识图谱的关键部分。在 PKG 工具中，这些功能要么完全缺失，要么尚处于萌芽阶段，某些工具通过变通方法提供支持。
An important aspect of open and enterprise knowledge graphs, that is currently almost entirely absent in PKG tools, is support for semantics. Controlled vocabularies with formal definitions and typed relationships/links with explicit semantics are a key part of open knowledge graphs. In PKG tools these are either entirely missing or embryonic, supported via workarounds in some tools.
Solid 是这一方向的一个有前途的方法。Solid 是一个网络去中心化项目，由万维网发明者 Tim Berners-Lee 爵士领导，最初在麻省理工学院合作开发。该项目“旨在通过开发一个完全去中心化且完全由用户控制的链接数据应用程序平台，从根本上改变当今 Web 应用程序的工作方式，实现真正的数据所有权以及更好的隐私保护”[7]。
One promising approach in this direction is Solid. Solid is a web decentralization project led by Sir Tim Berners-Lee, the inventor of the World Wide Web, originally developed collaboratively at MIT. The project “aims to radically change the way Web applications work today, resulting in true data ownership as well as improved privacy” [7] by developing a platform for linked data applications that are completely decentralized and fully under users’ control.
我们认为 Solid 与 PKG 相关的原因是它基于链接数据/RDF 数据模型。这意味着它支持图形数据模型。从事 Solid 工作的人们也持这种观点。[8] 一家名为 Inrupt 的公司成立，旨在帮助建立一个商业生态系统来推动 Solid 的发展。尽管目前还没有基于 Solid 构建的商业应用程序，[9] 但它可以用作任何 PKG 工具的基础。
The reason we consider Solid relevant for PKGs is that it is based on a linked data/RDF data model. That means it supports the graph data model. The people who work on Solid also share this view. [8] A company called Inrupt has been founded to help build a commercial ecosystem to fuel Solid. Even though there are no commercially available applications built on Solid yet, [9] it could be used as a substrate for any PKG tool.
另一个有趣的方向，可以看作是 PKG 的演变，即朝着集体/协作使用 PKG 工具的方向发展。乍一看，这似乎是一个悖论，但仔细研究后，可能就不那么矛盾了。
Another interesting direction, which could be seen as an evolution of PKGs, is towards collective/collaborative use of PKG tools. That may initially seem like a paradox, however, on closer examination it may be less paradoxical.
虽然当今 PKG 工具的主要方面是通过笔记记录提供数据收集、生成和注释的个人视角，但协作和互操作性也是一个重要方面。毕竟，知识工作不是在真空中进行的。
While the primary aspect of PKG tools today is to serve a personal view of data collection, generation and annotation via note-taking, collaboration and interoperability is also an important aspect. After all, knowledge work does not happen in a vacuum.
许多 PKG 工具已经提供协作功能，而其他工具则正在朝这个方向发展。协作编写文档的能力可能被视为唾手可得的成果。集体知识创造和意义建构等功能在 PKG 工具领域中还比较遥远，但也提供了更大的潜在收益。我们希望看到 PKG 生态系统不断发展，使用户能够从个人知识图谱转向人际知识图谱，类似于 Agora（Ivanec，本卷）。
Many PKG tools already offer collaborative features, while others are moving in that direction. The ability to collaboratively author documents may be seen as low-hanging fruit. Features such as collective knowledge creation and sense-making are more distant on the PKG tool horizon, but also offer greater potential yield. We would like to see the PKG ecosystem evolve, enabling users to go from personal to interpersonal knowledge graphs, similar to Agora (Ivanec, this volume).
然而，可以说，要实现基于 PKG 工具的集体智慧、感知和决策的愿景，我们还有很长的路要走。目前，PKG 生态系统生机勃勃，但在互操作性和语义基础方面仍有很多不足之处。我们仍然希望，随着生态系统的发展，从知识体系和共享经验中学习，这些方面将会取得进展。
Arguably, however, we have a long way to go to realize the vision of collective intelligence, sense-making and decision-making based on PKG tools. The PKG ecosystem in its current state is lively and blooming but leaves a lot to be desired in terms of interoperability and semantic grounding. We remain hopeful that progress on those fronts will be made as the ecosystem evolves, learning from a body of knowledge and shared experience.
笔记
Notes
[1]例如：https://research.gigaom.com/report/gigaom-radar-for-graph-databases/
[1] For example: https://research.gigaom.com/report/gigaom-radar-for-graph-databases/
[2] 区块也被定义为可互操作的组件，在流行的内容管理系统中得到广泛使用。还有一个基于区块的应用程序的开放标准，称为区块协议 https://blockprotocol.org/。一些工具还为区块使用单独的标识符，使其可引用。
[2] Blocks are also defined as interoperable components and are widely used in popular content management systems. There is also an open standard for block-based apps called the Block Protocol https://blockprotocol.org/. Some tools also use separate identifiers for blocks, making them referenceable.
[3] 抽认卡 (也称为索引卡) 是一种两面都载有信息的卡片，可以用来辅助记忆。
[3] A flashcard (also known as an index card) is a card bearing information on both sides, which can be used as an aid in memorization.
[4] 所有评估工具的资金和人力细节均由 Crunchbase 和 LinkedIn 提供
[4] Funding and workforce specifics for all evaluated tools were sourced by Crunchbase and LinkedIn
[5] https://www.youtube.com/live/TkNTuFF2t-c
[5] https://www.youtube.com/live/TkNTuFF2t-c
[6] https://www.theinformation.com/articles/a-200-million-seed-valuation-for-roam-shows-investor-frenzy-for-note-takeing-apps
[6] https://www.theinformation.com/articles/a-200-million-seed-valuation-for-roam-shows-investor-frenzy-for-note-taking-apps
[7] https://www.csail.mit.edu/research/solid-social-linked-data
[7] https://www.csail.mit.edu/research/solid-social-linked-data
[8] https://www.youtube.com/watch?v=2EP35HO2HVQ
[8] https://www.youtube.com/watch?v=2EP35HO2HVQ
[9] 目前，英国广播公司（BBC）已经开发出一个原型，弗兰德斯政府也在持续努力
[9] At the moment, there is a prototype by BBC and an ongoing effort by the Flemish Government
概念、实践和愿景
Concepts, practices and visions
第四章
Chapter 4
知识的无意部分在PKG中起着决定性的作用
The decisive role of the unintentional part of knowledge in PKGs
法布里斯·加莱
有谁没有被流行的 PKG（个人知识图谱）的视觉表现的美感和复杂性所震撼和吸引呢？[1] 经过几个月的认真记录，一个由数千个关系和节点组成的看似无法解开的纠缠出现了，并且似乎像一个活生生的生命一样成长。
Who hasn’t been struck and seduced by the beauty and complexity of the visual representation of a trendy PKG (personal knowledge graph)? [1] After a few months of conscientious note-taking, a seemingly inextricable tangle of thousands of relations and nodes appears and seems to grow like a living being.
乍一看，PKG 的视觉结构与任何 KG（知识图谱）的结构相似，PKG 的使用使其更加大众化。它是一组连接在网络中的元素，元素是图的节点，关系是边。但如果我们仔细观察，PKG 中建立的关系与 KG 标准的质量截然不同，并且呈现出的复杂性不仅仅来自于它们的数量。
At first sight, the visual structure of a PKG is like that of any KG (knowledge graph), whose use PKGs democratize. It is a set of elements connected in a network, the elements being the nodes of the graph and the relations the edges. But if we take a closer look, the relations established in a PKG are of very different quality from the KG standards and present a complexity that does not only come from their number.
在本章中，我们将研究 PKG 中关系的建立方式，特别是定义关系的意向性程度。我们将看到我们可以在多大程度上利用它们来丰富我们的知识，以及它们的泛滥会带来哪些问题。在简要概述了有助于我们认识和探索 PKG 中无意识关系的实践之后，我们将看到这种系统如何让我们超越线性写作的限制并回应苏格拉底对它的批评。
In this chapter, we will examine the way relations are established in PKGs, in particular the degree of intentionality involved in defining them. We will see to what extent we can take advantage of them to enrich our knowledge and what problems are caused by their proliferation. After a brief overview of the practices that help us become aware of and explore the unintentional relations in a PKG, we will see how such a system allows us to go beyond the limits of linear writing and responds to Socrates’s criticisms of it.
超越意向关系的 PKG 隐藏的复杂性
The hidden complexity of PKGs, beyond intentional relations
在 PKG 中建立关系是直观的，并且不受正式限制
Establishing relations in a PKG is intuitive and free of formal constraints
近年来，最流行的 PKG 逐渐为大众所知，它们被呈现为笔记应用程序，允许轻松链接任何类型的信息，而无需任何预先建立的架构。你所要做的就是写下来：无需问自己要创建什么类型的实体、它的属性是什么、它的授权关系是什么等。
The most popular PKGs, which have become known to the general public in recent years, are presented as note-taking applications that allow any type of information to be easily linked without any preestablished schema. All you have to do is to write: there is no need to ask yourself what type of entity you are creating, what are its properties, what are its authorized relations, etc.
用户不一定知道底层存储是图形数据库（或其模拟），也不一定知道图形数据库是什么。他们通过浏览信息并可视化简化模型，直观地了解他们在信息之间编织的关系的图形结构。无需了解图论或 KG 领域的现有标准。
Users are not necessarily aware that the underlying storage is a graph database (or a simulation thereof), or what a graph database is. They have the intuition of the graph structure of the relations they weave between the information by browsing it and visualizing a simplified model. No knowledge of graph theory or existing standards in the KG domain is required.
PKG 流行的原因之一是可以轻松即时创建关系（例如，通过按“[[”或“@”或“#”，或通过拖放）。因此，大量关系被建立，并且通常没有精确定义。此外，PKG 通常不提供专用接口来定义关系。关系保持隐式，并由根据上下文建立关系的人理解。
One of the reasons for the popularity of PKGs is the ease with which relations can be created on the fly (e.g., by pressing “[[” or “@” or “#”, or with drag and drop). As a result, relations are established in large numbers and usually without being precisely defined. Moreover, PKGs do not generally offer a dedicated interface to define a relation. Relations remain implicit and understood by the one who establishes them according to the context.
由于使用简单，自发形成的网络的复杂性很快变得令人眼花缭乱。简单结构和复合结构都可以轻松处理。简单结构是与明确定义的真实实体（如人、书或地点）相对应的节点。复合结构是代表多个真实实体之间关系的节点，例如判断、论点或列表。
Because of this simplicity of use, the complexity of the spontaneously formed network quickly becomes dizzying. Both simple and compound structures can be handled without friction. Simple structures are nodes corresponding to a well-defined real entity such as a person, a book, or a place. Compound structures are nodes such as a judgment, an argument, or a list which represent relations between several real entities.
例如，一个人对另一个判断做出判断，从而建立复合关系，其中一个节点本身就是一个关系。如果必须遵守严格的本体论和严格的构造规则，那么这一切都将非常困难。与知识图谱构造的严谨性相比，知识图谱的易用性（以及围绕其中所捕获思想的模糊性）类似于日常语言的自发性与形式逻辑的严谨性之间的对比。
For example, one makes a judgment about another judgment, thus establishing a compound relation, where one of the nodes is itself a relation. All this would be very difficult if one had to respect a strict ontology and rigorous construction rules. The ease of the PKG – but also the ambiguity that surrounds the ideas that are captured in it – compared to the rigor in the construction of KGs is analogous to the spontaneity of everyday language compared to the rigor of formal logic.
根据意向性程度，我们可以区分三类关系
We can distinguish three types of relations according to their degree of intentionality
1. 意向关系
1. Intentional relations
首先，存在有意、刻意的关系，它们记录给定的知识，因为这是知识图谱的目的。例如，为了表达“是……的来源”这个关系，我们可以在包含引文的页面顶部插入节点 Source:: https://... [2]。
First, there are intentional, deliberate relations, which record given knowledge, since this is the purpose of a KG. For example, to express the relation “is a source of”, we can insert the node Source:: https://... [2] on top of a page containing a quote.
我们可以使用相同的 Source:: 节点来指示给定的引言是给定主张的来源。当我们将一条信息作为证据来支持主张、将另一条信息作为证据来定义概念等时，情况也是一样的。
We can use the same Source:: node to indicate that a given quote is the source of a given claim. It is the same when we present a piece of information as evidence that [[supports]] a claim, another that [[defines]] a concept, etc.
这些关系与标准 KG 中的关系类似，即使它们的定义可能不那么严格。在上例中，相同的关键字“Source”指的是不同类型的关系，但上下文足以避免混淆。
These relations are similar to those found in standard KGs, even if their definition may be less rigorous. In the previous example the same keyword “Source” refers to different types of relations, but the context is sufficient to avoid confusion.
2.半意向关系
2. Semi-intentional relations
在 PKG 中，我们通常会在不严格思考其性质的情况下创建关系。关系是在图形数据库中建立的，我们知道要这样做，但我们是在运行中这样做的。
Commonly in PKGs, we create relations without rigorously reflecting on their nature. A relation is established in the graph database and we are aware of doing it, but we do it on the fly.
通常，这意味着创建一个“指针”，即使用超文本链接而不是特定关系来连接到图中的另一个节点。在这里，我们可以说隐式或部分有意的关系。我们将使用半有意的表达来将它们与完全有意和无意的关系区分开来。
Typically, this means creating a “pointer”, a bridge to another node of the graph, using say a hypertext link, rather than a specific relation. Here, we could speak of implicit or partially intentional relations. We will use the expression semi-intentional to distinguish them from fully intentional and unintentional relations.
自然语言本身就是一个强大的工具，用于关联或表达原子或复合思想之间的关系。语言的工作方式往往非常模糊，尽管我们付出了所有努力，也很难明确表达。这种对关系严谨性的有限要求也是流畅使用语言和创造力的条件。
Natural language is in itself a formidable tool for relating or expressing relations between atomic or compound ideas. Language works in a way that is often very fuzzy and difficult to make explicit despite all our efforts. This limited requirement for the rigor of relations is also a condition for fluid use of language and creativity.
类似地，PKG 系统为自然语言带来了连接的可能性。经过一点练习，这些连接就会变得像自然语言连接一样“自然”。在图中动态创建的关系就像添加到话语中的第三维度。如果第一个是简单的归因（句子），第二个是线性清晰的话语，第三个则以非线性的方式跨越话语。
In a similar way, PKG systems bring connection possibilities to natural language. With a little practice, these become as “natural” as natural language connections. The relations created on the fly within the graph are like a third dimension added to the discourse. If the first is simple attribution (the sentence), the second is linear articulated discourse, and the third crosses discourse in a nonlinear way.
假设我们在页面 [[happiness]] 中写下这些内容：
Let’s say we write this in the page [[happiness]]:
图 4.1 Roam Research 中的半意向关系示例
Figure 4.1 Example of semi-intentional relations in Roam Research
自然语言在这里设置了很多关系。首先是句子内的关系，例如归因和连词。然后设置句子之间的关系，形成一个有前提和结论的逻辑推理，而不必将其形式化。通过让一些关键词引用其他页面，如这里的“自然”、“必要”、“欲望”、“友谊”或“财富”，在它们每个与“幸福”节点之间建立链接。
Natural language here sets a lot of relations. First, relations within a sentence, such as attribution and conjunction. Then, relations between sentences are set, forming a logical reasoning with premises and a conclusion without necessarily formalizing it. By making some keywords references to other pages, like “natural”, “necessary”, “desire”, “friendship” or “wealth” here, a link is created between each of them and the “happiness” node.
当一个词被当作参考时，它就会打开一条通往该参考的路径。它可以包含支持或完成所说内容的信息。例如，[[友谊]]页面可能引用了亚里士多德关于友谊的理论。这也使其成为幸福和美德的重要支撑。或者它可能包含不同甚至误导的信息。例如，伊壁鸠鲁在这里理解的自然意义与现代生物学意义上的自然意义并不一致。
When a word is taken as a reference, it opens a path to that reference. It can contain information that supports or completes what is said. For example the [[friendship]] page may reference Aristotle’s theory on friendship. That also makes it an important support for happiness and virtue. Or it can contain information that differs or is even misleading. For example, what is natural in the sense that Epicurus understands here does not correspond to what is understood as natural in the modern biological sense.
此外，通过引用这些术语，创建了“友谊”和“自然”之间的关系。例如，我们可以从页面“友谊”中看到这一点：它与“幸福”有直接关系，与“自然”有间接关系。这是因为“友谊”在子块中引用了“幸福”页面上对“自然”的引用。
Moreover, by referencing these terms, a relation between “friendship” and “natural” has been created. We can see it, for example, from the page “friendship”: there is a direct relation to “happiness” and an indirect one, among others, to “natural”. That’s because “friendship” has been referenced in a child block to the reference to “natural” on the page “happiness”.
自然语言的语法使这种关系变得清晰。然而，在图中，它没有被指定，仍然很模糊。事实上，它以与“财富”和“自然”之间的关系相同的方式记录，它们是相反的，以及“欲望”和（对）“友谊”的关系，这是一种类与子类的关系。
The syntax in natural language makes this relation clear. In the graph, however, it is not specified and remains vague. In fact, it is recorded in the same way as the relation between “wealth” and “natural”, which is the opposite, and “desire” and (desire for) “friendship”, which is a class to subclass relation.
图 4.2 半有意向关系：链接到“友谊”页面的页面
Figure 4.2 Semi-intentional relation: pages linked to “friendship” page
为什么引用某些术语而不是其他术语？这些引用的期望是什么？大多数时候，我们凭直觉建立这些关系，通过“感觉”某些链接值得建立并且以后可能会有用。这些半有意的关系基于经验知识，而不是深思熟虑的知识或关系类型学。
Why are certain terms referenced instead of others? What is expected from these references? Most of the time we establish those relations intuitively, by “feeling” that certain links are worth setting up and may be useful later. These semi-intentional relations are based on empirical know-how, rather than thoughtful knowledge or a typology of relations.
3. 无意的关系
3. Unintentional relations
最后，由于媒介的工作方式，存在大量无意识的关系。有些关系是在我们没有意识到的情况下记录下来的。自然语言已经在话语中散布的思想、信仰或感受之间编织了这种无意识的联系。通过统计模型或语义网络的图形可视化进行语义分析可以帮助揭示它们。
Finally, there is a vast set of unintentional relations, which results from the way the medium works. Some relations are recorded without our being aware of them. Natural language already weaves such unconscious links between ideas, beliefs or feelings scattered in a discourse. Semantic analysis, by means of statistical models, or graph visualizations of semantic networks, can help to reveal them.
在 PKG 系统中，无意链接对应于数据库引擎在后台记录的关系，而我们有意在图表的特定位置建立关系。这将直接决定数据在反向链接或搜索或查询中重新出现的方式。
In the case of PKG systems unintentional links correspond to relations recorded in the background by the database engine, while we intentionally establish a relation at a specific location of the graph. That will directly determine the way in which the data will resurface in the backlinks or in a search or a query.
例如，在将数据结构化为不同页面的 PKG 系统中，这些页面可能包含大量数据，每个用作节点的关键字除了更明确地链接到另一个元素之外，还将链接到该页面。让我们想象一下，在关于[[幸福]]的页面上，我们问对[[死亡]]的恐惧是否会阻碍幸福，我们接受了[[伊壁鸠鲁]]对死亡是一种[[邪恶]]的[[信仰]]的批评——根据他的说法，我们对此只能是[[无知]]。
For example, in a PKG system that structures its data in distinct pages, which may contain a large amount of data, each keyword used as a node will be linked to the page in addition to being linked to another element more explicitly. Let’s imagine in the page on [[happiness]] we ask whether the fear of [[death]] prevents happiness and we take up [[Epicurus]]’s criticism of the [[belief]] that death is an [[evil]] – about which, according to him, we can only be in [[ignorance]].
在这里，我们链接了“信仰”和“幸福”、“邪恶”和“幸福”、“无知”和“幸福”（以及上例中提到的所有术语），但这些关系尚未确定，甚至不可取。这里反映的是幸福与死亡之间的关系。此外，如果 PKG 系统允许大纲结构，并且关于无知的陈述是关于邪恶的陈述的子陈述，则两者之间会建立关系。
Here we link the nodes “belief” and “happiness”, “evil” and “happiness”, “ignorance” and “happiness” (and all the terms mentioned in the previous example), without these relations being determined or even desirable. The relation reflected here is between happiness and death. If, moreover, the PKG system admits an outliner structure and the statement on ignorance is a child of the one on evil, a relation is made between the two.
同样，两个在话语中没有逻辑关系但存在于同一复合节点中的遥远概念都将链接到与其中一个相关的第三个概念。即使概念在初始上下文中没有特定关系，也会发生这种情况。
In the same way two distant concepts with no logical relation in the discourse, but present in the same compound node, will both be linked to a third concept which would be related to one of them. That happens even if the concepts have no particular relation in the initial context.
如果这些无意的关系在 PKG 中很丰富，那么它们在集体图中会更加丰富，就像这本书的图表版本一样：几位作者会在不知情的情况下使用相同的关键词，因此他们的反思会相互影响。
If these unintentional relations are abundant in a PKG, they will be even more so in a collective graph, as in this book in its graph version: several authors will use the same keywords without knowing it and so their reflections will feed each other.
通过今天建立的联系，我们忘记了过去建立的大部分联系。这些联系为过去的笔记恢复了“生命”。所以我们今天建立的联系也是一种瓶中信，也许会被未来的自己找回。
By setting links today, we have forgotten a good part of the ones we have established in the past. These links restore “life” to the past notes. So the links we create today are also a kind of message in a bottle which will perhaps be recovered by a future self.
更甚的是，如果系统能够自动引用与图谱节点相对应的任何文本，甚至根据识别图中其他数据链接的算法添加标签，那么无意的关系就会成倍增加。这样的工具对 PKG 的质量和可用性是有益还是有害？
Even more so, a system that automatically references any piece of text that corresponds to a node of the graph, or even that adds tags according to an algorithm that identifies links to other data in the graph, would multiply the unintentional relations. Would such a tool be useful or harmful to the quality and usability of the PKG?
因此，普通人在 PKG 中建立的关系比在图表的视觉表示中观察到的关系要多得多、复杂得多。它们的复杂性几乎不可能明确表达出来。半有意和无意链接的比例因所使用的 PKG 系统和人们喜欢的记录笔记的结构类型而异。
The relations established in a PKG by ordinary people are therefore much more numerous and complex than those that can be observed in the visual representation of the graph. Their complexity is such that it is almost impossible to make them explicit. The proportion of semi-intentional and unintentional links varies greatly depending on the PKG system used and the type of structure one prefers for recording notes.
类似 Roam 的 PKG 无疑有助于关系的倍增。这也因人的严谨程度而异。有些用户会明确大多数关系，避免在运行中建立任何关系，而另一些用户则喜欢在运行中建立关系，而不会在当时过多考虑它们，以遵循他们当前的思维和直觉，然后再回来添加结构。Roam 的方法以每日笔记页面为中心，而不是以主题或文件夹为中心，在很大程度上鼓励了第二种方法。
Roam-like PKGs certainly facilitate the multiplication of relations. It also varies according to one’s rigor. Some users make most relations explicit and avoid any relations on the fly, others prefer to make relations on the fly, without thinking too much about them at the time, to follow the flow of their current thinking and intuitions and come back to them later to add structure. Roam, with its approach centered on daily note page rather than on topics or folders, largely encourages the second approach.
我们是否应该将 PKG 中这些半故意和无意关系的激增视为混乱和噪音的来源，担心它们最终会使 PKG 无法使用，并倾向于尽可能减少它们？或者这些关系可能是有用的？
Should we consider the proliferation of these semi- and unintentional relations in the PKG essentially as a source of disorder and noise, fear that they will eventually make the PKG unusable and tend to reduce them as much as possible? Or could those relations be useful?
我们是否应该考虑半故意和无意的关系，或者它们可以忽略不计？
Should we take semi- and unintentional relations into account or are they negligible?
大脑和 PKM（包括 PKG）之间的类比现在已很常见。该图可视为第二个大脑。它可以首次暗示半意向或无意向关系的重要性。正如我们承认意识和意向操作仅对应于我们大脑中发生的事情的一小部分一样，PKG 中的意向关系仅代表记录的关系的一小部分。
The analogy between the brain and the PKM (including the PKGs) is now commonplace. The graph can be seen as a second brain. It can give a first hint of the importance of semi- or unintentional relations. In the same way that we acknowledge that consciousness and intentional operations correspond to only a small part of what happens in our brain, the intentional relations in a PKG represent only a small part of the relations recorded.
类比到此为止，因为很明显，大脑的复杂程度是另一个数量级，其动态性和可塑性在当前的知识图谱中是没有对应物的。最后，神经元连接的地位与知识图谱中实体之间的关系完全不同。
The analogy stops there because it is obvious that a brain has a degree of complexity of another order of magnitude and its dynamism and plasticity have no equivalent in current PKGs. Finally, neuronal connections have a completely different status than the relations between entities in a KG.
然而，正如对某些大脑机制的理解可以解释思想、偏见或令人惊讶的行为的关联一样，我们可以以类似的方式假设，我们 PKG 中某些关系的暴露可以阐明我们的知识状态，并以某种方式巩固或增加它。
However, just as the understanding of certain cerebral mechanisms can explain associations of ideas, biases or surprising behaviors, we can suppose that in an analogous way, the exposure of certain relations in our PKG could shed light on the state of our knowledge and, in a certain way, consolidate or increase it.
有些关系很琐碎
Some relations are trivial
我们必须承认，许多关系实际上微不足道或价值不大。有些关系源于同音异义词的存在，另一些关系则是轶事，源于人们注意到几组没有相关逻辑联系的想法的上下文。在同一页上做大量笔记或将整篇文章或书籍复制到图表中可能会使图表充斥着无法真正利用的关系。
We have to admit that many relations are in fact trivial or of little value. Some result from the existence of homonyms, others are anecdotal and result from the context where several sets of ideas without relevant logical links have been noted. Taking a lot of notes on the same page or copying entire articles or books into the graph can saturate it with relations that are not really exploitable.
为了避免噪音淹没相关信息，我们可以采用一些良好的组织实践：
To avoid having noise drowning out relevant information, we can adopt some good organizational practices:
在写草稿时创建一些链接，
create few links when writing drafts,
将草稿或版本移至另一个图表，或将其归档，
move drafts or versions to another graph, or archive them,
逐渐建立最常用的链接类型的约定，
gradually set up conventions for the most frequently used types of links,
以分析性（或原子性）的方式写笔记，即将不同的想法分成不同的块或页面等。
write notes in an analytical (or atomic) way, i.e., separate distinct ideas in different blocks or pages, etc.
半意向关系至少在元认知层面上具有指导意义
Semi-intentional relations are at least instructive on the metacognitive level
对于半有意向关系，用户直观地知道他们正在编织的关系，而无需明确说明。PKG 不会记录比已建立关系更多的信息。知识更多地存在于用户的头脑中，而不是图表中。但 PKG 突出显示此关系（例如，通过使其重新出现在反向链接部分，或仅使用不同颜色的单词将其标识为图表中的节点）这一事实会邀请用户考虑此链接。
In the case of a semi-intentional relation, the user knows intuitively the relation they are weaving without making it explicit. The PKG does not record more information than the fact that a relation has been established. The knowledge is more in the user’s head than in the graph. But the fact that the PKG highlights this relation (e.g., by making it reappear in the backlinks section, or simply with a different colored word to identify it as a node in the graph) invites the user to take this link into consideration.
现在，虽然我们知道我们已经建立了一种关系，并且我们将其视为我们知识的一部分，但我们不一定知道如何像我们认为的那样很好地解释它。例如，即使我们知道自行车有一条链条，它可以驱动车轮的运动，但我们不一定能够详细解释这种关系。
Now, although we know we have established a relation and we recognize it as part of our knowledge, we do not necessarily know how to explain it as well as we think we could. For example, even though we know that a bicycle has a chain, which drives the movement of the wheel, we are not necessarily able to explain this relation in detail.
我特意举这个例子，因为这个想法可以通过一个著名的实验来说明，这个实验强调了解释深度的错觉。[3] 在这个实验中，人们被问到他们是否认为他们可以解释自行车的工作原理。大多数人认为他们可以，但当被要求这样做时，许多人意识到他们并不确切知道自行车是如何工作的（以至于画出完全没有功能和滑稽的自行车）。
I take this example deliberately, since this idea can be illustrated by a famous experiment that highlighted the illusion of explanatory depth. [3] In this experiment people are asked if they think they can explain how a bicycle works. Most people think they can but when asked to do it many realize that they don’t know exactly how a bicycle works (to the point of drawing totally non-functional and funny bikes).
同样，通过在 PKG 中即时建立关系，我们可能认为我们知道它是什么，但我们可能经常很难准确解释它。这种关系在我们的 PKG 中的存在表明我们在特定时刻对其价值有一些直觉。这至少是一种元认知迹象。
Similarly, by establishing a relation on the fly in our PKG, we probably think we know what it is, but we would probably often have a hard time explaining it precisely. The existence of this relation in our PKG points to the fact that we had some intuition of its value at a given moment. It is at least a metacognitive indication.
关系表明我们相信我们知道关系由什么组成，或者为什么建立关系，或者至少它是有意义的。回想起来，这种联系的可见性就像是邀请我们质疑其性质并测试我们的解释能力。半意向关系是一种潜在的意向关系，如果要对相关对象进行更严肃的知识工作，就需要明确和定义这种关系。
The relation indicates that we believe we know what the relation consists of, or why it has been established, or at least that it is meaningful. In retrospect, the visibility of the link is like an invitation to question its nature and to test our capacity for explanation. A semi-intentional relation is a potential intentional relation, which would need to be made explicit and defined if more serious knowledge work were to take place on the object in question.
PKG 的一大优势是其灵活性极高。一个关系可以很容易地重塑和重组为几个精确的关系。它是一种严格的 KG 的永久草案，从中逐渐出现更明确的结构。这种灵活性的另一方面是，现有关系的很大一部分仍然存在一定程度的模糊性。然而，当需要时，总是可以通过指定其性质和所连接元素的类型来明确关系。
One of the great advantages of PKGs is their great flexibility. A relation can be easily reshaped and recomposed into several precise relations. It is a sort of permanent draft of a rigorous KG, from which better defined structures emerge gradually. The flip side of this flexibility is that there is still some degree of ambiguity in a significant portion of existing relations. It is, however, always possible to make a relation explicit by specifying its nature and the type of the elements connected, when the need arises.
但如果半意向关系需要更严格地重新考虑，一般来说就不会发生这种情况。大多数半意向关系不会被重新考虑，因为它们的价值主要在于直觉的实验记录，然后可以进行有指导意义的探索。
But if a semi-intentional relation is a call for a more rigorous reconsideration of the relation, in general it will not happen. Most semi-intentional relations will not be reconsidered because their value is primarily that of the experimental recording of intuition, which then allows for instructive exploration.
PKG 中可以区分出两个“层”。一个层是结构化的，其中实体和关系定义得足够精确，可以快速搜索信息。但我们还有第二个更模糊的层，它通过联想、直觉和试验发挥作用，有利于偶然发现和探索。[4] 从第二层开始，可能会出现新的结构或第一层的修改，其中记录并确立了“构成知识”。问题是，这两个层是否应该共存于同一个 PKG 中，或者是否有必要为此目的维护两个不同的图表。
One may distinguish two “layers” in a PKG. One layer is structured, where entities and relations are defined precisely enough to allow for a quick search for information. But we also have a second, more nebulous layer, which functions by association, intuition, and trials, which favor serendipity and exploration. [4] From this second layer, new structures or a modification of the first layer may emerge, where “constituted knowledge” is recorded and well established. The question is whether it is desirable for these two layers to coexist in the same PKG, or whether it would be necessary to maintain two distinct graphs for this purpose.
这个问题的答案取决于人们重视半意向关系和无意向关系在知识形成中所起的作用。我们认为，它们有助于丰富知识，并从不同角度审视知识。特别是，它们使我们能够将知识从我们倾向于将其锁定的“孤岛”中解放出来，正如我们现在将展示的那样。因此，重要的是将最结构化和最直观的两个层次保持在同一个 PKG 中，这样构成的知识就不会与思维流隔绝。
The answer to this question depends on the importance one attaches to the role played by semi- and unintentional relations in the formation of knowledge. We argue that they contribute to the enrichment of knowledge, and to its interrogation from different perspectives. In particular, they allow us to get our knowledge out of the “silos” where we tend to lock it up, as we will now show. It is therefore important to maintain the two layers, the most structured and the most intuitive, in the same PKG, so the constituted knowledge is not cut off from the thought flow.
克服人为知识分割的关系是宝贵的
Relations that overcome the artificial partitioning of knowledge are precious
在两个通常孤立的领域之间建立联系的关系尤其有价值。PKG 的优势之一当然是它们使我们能够在所有知识和经验领域之间建立桥梁，从而提供原创观点和更好的综合知识。知识的价值尤其取决于它是否具有启发性，是否能带来新的见解。在建立关系时，这些见解是无法察觉的，或者只是勉强察觉到的。
Relations that weave a link between two commonly isolated domains are especially valuable. It is, of course, one of the strengths of PKGs that they allow us to build bridges between all the fields of our knowledge and experience, thus offering original perspectives and better integrated knowledge. The value of knowledge is measured, in particular, by the fact that it is enlightening, that it allows new insights. Insights that are not, or only barely, perceived at the time the relation is established.
除了简单的分类之外，想象一下这样的场景：我们不仅将一篇文章链接到“文章”、“认知心理学”和“解释”（例如，对于前一篇文章），而且还给它贴上“幻觉”、“偏见”、“大脑”、“信仰”、“无知”、“自命不凡”、“技术”等标签。然后，我们不仅可以将文章链接到认知心理学领域，还可以链接到例如人工智能或工作相关冲突领域。
Beyond simple classification, imagine a scenario where we do not only link an article to “article”, “cognitive psychology” and “explanation” (e.g., for the previous article), but we also tag it with “illusion”, “bias”, “brain”, “belief”, “ignorance”, “pretention”, “technology”, etc. We can then link the article not only to the field of cognitive psychology, but also, for example, to that of AI or work-related conflict.
与人工智能的联系可能指的是机器学习的可解释性问题。与工作冲突的联系可能指的是对无法解释某事的同事的不经意批评，认为做出判断的部分可以做得更好。
The link to AI may refer to the machine learning explainability problem. The link to work-related conflict may refer to careless criticism aimed at colleagues who are incapable of explaining something, assuming the part passing judgment could do it better.
将这个例子推广开来，可以指向更广泛的自我认知问题，即评估自己的知识。这不能仅通过内省来实现：人们必须尝试解释或绘制才能意识到自己无法做到这一点（例如，知道或不知道自行车如何工作）。这些关系中的每一个，以及每个领域之间的关系仍然很模糊，但人们认为，深入研究其中一个关系将为其他关系带来有趣的见解。
Generalizing this example points towards the broader issue of self-knowledge, i.e., assessing one’s own knowledge. This cannot be done by introspection alone: one must try to explain or to draw to realize that one is incapable of doing so (as in the example of knowing or not knowing how a bicycle works). Each of these relations, and the relations between each of these domains, remains vague but one perceives that digging into one of them will bring interesting insights into the others.
无意关系的重要性更难把握
The importance of unintentional relations is more difficult to grasp
有人可能会天真地认为，通过增加连接，可以增加获得原创见解的机会。[5] 然而，“混乱”的风险非常明显，太多的关系会使系统饱和。但必须承认，理论上，任何关系都有可能在某一天变得有启发性。我们无法知道它们会发生什么。每次我们将一条新信息链接到现有节点时，我们都会以类似于神经网络演化的方式间接修改先前关系的值。
One could naively say that by multiplying the connections, one multiplies the chances of original insight. [5] The risk of “mess” is however very present and too many relations can saturate the system. But it must be admitted that, in theory, any relation may turn out to be enlightening one day. We cannot know what will happen to them. Each time we link a new piece of information to an existing node, we indirectly modify the value of the previous relations in a way analogous to the evolution of a neural network.
这是因为每个节点不再具有完全相同的“权重”、相同的属性或相同的含义。例如，如果我们将图 4.1 中用来描述某些欲望的“自然”节点与“极限”的概念联系起来，[6] 我们可能会增加我们对伊壁鸠鲁关于幸福的推理的理解。即使我们在思考伊壁鸠鲁时根本没有注意到这个想法，也会发生这种情况。在这样做的过程中，我们无意中揭示了一些以前的笔记。
That is because each node in question no longer has the exact same “weight”, the same attributes or the same meaning. For example, if we link the “natural” node used in figure 4.1 to characterize certain desires with the idea of “limit”, [6] we potentially increase our future understanding of Epicurus’s reasoning about happiness. That happens even though we did not notice this idea at all when thinking about Epicurus. In doing so, we unintentionally shed some light on previous notes.
很难确定无意关系在构成知识中起多大决定作用，因为我们不一定意识到我们依赖它们所形成的洞见。但我们可以说它发挥了重要作用。还有待考虑如何继续增加利用它的机会。
It is difficult to establish to what extent unintentional relations are decisive in constituted knowledge because we do not necessarily realize that we are relying on the insights they have made possible. But we can say it plays a significant role. It remains to consider how to proceed to increase the chances of taking advantage of it.
利用无意关系的一些策略
Some strategies to take advantage of unintentional relations
我们主要在这里对 PKG 中半或无意识关系的存在和扩散进行理论思考，这些半或无意识关系与标准 KG 不同。尽管如此，我们将简要提出将它们付诸实践的方法。这些只是一些仍然有限的想法，因为可能性在很大程度上取决于软件实现。
We are mainly conducting here a theoretical reflection on the presence and proliferation of semi- or unintentional relations in PKGs, which distinguish them from standard KGs. Nevertheless, we will briefly propose ways of putting them into practice. These are only a few ideas that remain limited because the possibilities depend strongly on the software implementation.
我们希望这些想法和新路径中的一些能够通过尚未发明的功能更容易地实现。我们必须谦虚地承认，我们仍处于 PKG 允许我们自发地增加关系并浏览它们的阶段，但尚未达到 PKG 指导和极大地促进这些关系的利用的阶段。
We hope some of these ideas and new paths can be followed more easily using features yet to be invented. We must humbly acknowledge that we are still at the stage where PKGs allow us to spontaneously multiply relations and browse them, but not yet at the stage where PKGs guide and greatly facilitate the exploitation of these relations.
为了具体介绍，我们将首先快速介绍 Roam Research 的几个功能，这些功能可以直观地浏览我们图中逐渐形成的所有关系。Roam Research 是作者最熟悉的 PKG 系统，但在许多竞争应用程序中都可以找到等效系统，例如 Logseq、RemNote、Obsidian 或 Tana。这是一个快速概述，主要针对那些不使用此类 PKG 系统的人，该系统并不声称是详尽无遗或原创的。然后，我们将简要介绍可以基于这些功能或类似功能更通用的方法。
To present things concretely, we will first quickly refer to several features of Roam Research that enable intuitive browsing of all relations gradually formed in our graph. Roam Research is the PKG system with which the author is most familiar, but equivalents can be found in many competing applications such as Logseq, RemNote, Obsidian or Tana. This is a quick overview mainly intended for those who do not use this type of PKG system, which does not claim to be exhaustive or original. Then, we will briefly suggest more general approaches which could be based on these features or similar ones.
漫游功能可利用意外关系
Roam features to leverage unintended relations
链接引用或反向链接
Linked references or Backlinks
反向链接是 Roam Research 推广的主要功能之一，现在大多数 PKG 系统都有此功能。当您从页面 A 链接到页面 B 时，会自动创建从 B 到 A 的反向链接，该链接显示在页面 B 底部的“链接引用”标题下。当您创建初始链接时，无法理解此反向链接的价值，因为您的视角来自 A。
Backlinks are one of the main features popularized by Roam Research and now found in most PKG systems. When you link from page A to page B, the reciprocal link from B to A is automatically created and visible at the bottom of page B, under the “Linked references” heading. The value of this backlink is not understood when you create the initial link, since your perspective is from A.
但是，当您稍后访问节点 B 时，您的心态以及您关心的想法和问题可能会揭示反向链接中的意外价值。以“幸福”和“死亡”之间的联系为例：当我们考虑幸福和考虑死亡时，与当我们考虑死亡和考虑幸福时，会产生不同的想法。此外，同一反向链接列表中不同元素之间出现的关系是无法预测的，并且可以产生新的想法。
But your state of mind and the ideas and problems that concern you when you later visit node B can reveal unexpected value in the backlink. To take our example on the link between “happiness” and “death”: different thoughts occur when we think about happiness and consider death versus when we think about death and consider happiness. Furthermore, the relations that appear between the different elements of the same list of backlinks cannot be anticipated and can give rise to new ideas.
未链接的引用
Unlinked references
未链接的引用是一种从根本上来说无意的反向链接类型，它仅依赖于页面标题和一段文本之间的词汇身份。如果用户认为相关，则邀请他们将图表中两个节点之间的链接设为“官方”。
Unlinked references are a fundamentally unintentional type of backlink and rely solely on the lexical identity between the title of the page and a piece of text. The user is invited to make the link between two nodes of the graph “official”, if they consider it relevant.
可以利用此功能，创建页面，标题为希望成为图表新重要节点的短语，从而将图表中某些元素之间的潜在关系转化为实际关系。例如，我们创建一个“我需要”页面，并让它在眨眼间成为了解和列出我的需求的新重要节点。
It is possible to play with this feature by creating pages with the title of a phrase that one would like to see become a new significant node of the graph, and thus transform potential relations between certain elements of the graph into actual relations. For example, we create a page “I need” and make it, in the blink of an eye, a new significant node to know and list my needs.
过滤器（用于页面或链接引用）
Filters (for page or linked references)
Roam 用户体验中一个被低估的功能是过滤链接引用的能力，它只保留包含选定引用集的链接，或者相反，不包含某些引用的链接。在这里，无意链接最有用的部分不是过滤功能（当有大量链接时它可能很有用），而是用户界面。
A feature that is underrated in the Roam user experience is the ability to filter linked references, retaining only links that include a selected set of references, or on the contrary that do not include certain references. Here, the most informative part of unintentional links is not so much in the filter function (useful as it may be when there are a lot of links) but rather in the user interface.
用户界面显示链接到当前页面的节点列表，按提及次数降序排列。打开“链接引用”过滤器时，此列表本身可快速概览链接到此页面的所有节点，打开页面过滤器（位于页面顶部）时，可快速概览从此页面到其他节点的所有节点。
The user interface presents a list of nodes linked to the current page, sorted by decreasing number of mentions. This list alone gives a quick overview of all the nodes linked to this page when opening the “Linked references” filter, and of all the nodes from this page to other nodes when opening the page filter (at the top of the page).
这样，我们就得到了一个相当于相邻节点和定向关系的图形表示，这里以列表的形式呈现，可以使用搜索框进行过滤。一旦列表变长，图形表示通常就不如这种线性表示方便。也就是说，只要我们不超出第一级关系。
In that way we have the equivalent of a graphical representation of adjacent nodes and oriented relations, presented here in the form of a list that can itself be filtered using a search box. As soon as the list gets longer, a graphical representation is often less convenient than such a linear presentation. That is, as long as we don’t go beyond the first level of relation.
块计数器，内联块计数器
Block counter, inline block counter
由于 Roam 中的关系是在块级别，而不仅仅是在页面级别，因此在某处引用的每个块都成为浏览其所有实例及其相关内容的入口点。原始块的右侧空白处有一个计数器，用于指示提及次数。单击此数字将显示提及列表及其上下文，可以快速浏览。
Since relations in Roam are at the block level and not just at the page level, each block that is referenced somewhere becomes an entry point to browse all its instances and everything related to it. The original block is accompanied by a counter, in the right margin, which indicates the number of mentions. A click on this number brings up the list of mentions, in their context, which can be quickly browsed.
块的每个引用还可以附带一个内联计数器（上标），这也使我们能够显示和浏览所有其他引用。这使我们能够在同一窗口中看到给定块的所有上下文（即所有重要但或多或少反映的关系集）。这意味着我们不仅可以比较和链接两个节点，还可以比较和链接不同的上下文。这些链接在图形可视化中根本不会出现。[7]
Each reference of a block can also be accompanied by an inline counter (in superscript), which also allows us to display and browse all the other references. This allows us to see, in the same window, all the contexts – thus all the sets of significant but more or less reflected relations – of a given block. This means we can not only compare and link two nodes, but also compare and link different contexts. These links do not appear at all in the graph visualization. [7]
查询
Queries
查询用于查询数据库并调出一组符合特定条件的元素，这些元素以逻辑方式组合在一起。这是任何数据库的基本功能，但由于存在许多未预料到的关系，我们期望的结果不一定是我们得到的结果。当涉及多个条件时，查询的语法会变得复杂。有一些扩展，特别是 RoamJS 的 Query Builder，[8] 提供了一个界面来简化查询的生成和利用。
Queries are used to interrogate the database and bring up a set of elements that meet certain criteria combined in a logical manner. This is a basic function of any database, but the result we expect is not necessarily the one we get because of the many unsuspected relations. The syntax of the queries can get complicated when multiple criteria are involved. There are extensions, in particular Query Builder by RoamJS, [8] that provide an interface to simplify generating and taking advantage of queries.
这个快速概览表明，可以使用方法来突出半故意或无意的关系。还有许多其他方法，并且该领域不断出现新的应用。可以设置非常复杂的工作流程，例如 Zettlekasten 方法的实施。我们只打算提供一些实践指南，让人们更加意识到这些关系的存在以及它们所构成的宝贵资源。
This quick overview shows that the means to highlight semi- or unintentional relations are available. There are many others and new applications are constantly innovating in this field. Very sophisticated workflows, such as the implementation of the Zettlekasten method, can be set up. We only intend to give a few guidelines for practices that will make people more aware of the existence of these relations and of the valuable ressource they constitute.
一些策略
Some strategies
只需浏览并享受！
Simply browse and enjoy!
在制定任何策略之前，要做的第一件事，也可能是最令人兴奋的事情，就是享受从各个方向浏览笔记的乐趣，让自己从图表上的一个点到另一个点。促进这一点的 PKG 已经做得很好了。有些人用漂亮的图表可视化来鼓励它。
Before any strategy, the first and possibly also the most exciting thing to do is simply to enjoy going through your notes in all directions, letting yourself be carried from one point to another on the graph. A PKG that facilitates this is already doing something excellent. Some encourage it with a nice visualization of the graph.
浏览硬盘上的文件夹和子文件夹显然不那么令人愉快和惊讶。通过浏览 PKG，你会发现一些翻阅旧笔记本的好奇心，你会对自己所写的内容感到震惊，在一页上停下来思考片刻或回顾当时的记忆，或者快速浏览几十页。
Browsing the folders and subfolders on a hard disk is clearly less pleasant and surprising. By browsing a PKG, one finds some of the curiosity of going through an old notebook, where one is astounded by what one has written, stopping on a page to reflect for a moment or to go back to the memories of the time, or quickly skimming through dozens of pages.
但在 PKG 中，路径总是不同的，你可以不遵循相同的路径穿过图一百次，只需享受出现的关系即可。这也许就是“漫游”的意思。它就像是对我们认知现实复杂性的冥想：它是关于接受我们将无法控制和指定其中形成的所有关系。
But in a PKG, the path is constantly different, you can cross the graph a hundred times without following the same path, and simply enjoy the relations that emerge. This is probably what “roam” is supposed to mean. It is like a meditation on the complexity of our cognitive reality: it is about accepting that we will not be able to control and specify all the relations that are formed in it.
定期审查
Scheduled reviews
更有序的方法是安排对图表不同区域的审查，以尽可能利用现有关系，避免被锁定在当前模式中，超越更线性的 PKM 工具。例如，您可以在每周审查中加入一次比主要笔记或项目更深入的游览，但同时也探索它们的链接或反向链接。
A more orderly way to proceed is to schedule reviews of different areas of the graph to take advantage of the existing relations as much as possible and avoid being locked in the current schema, going beyond a more linear PKM tool. You can, for example, include in your weekly review a tour that goes further than the main notes or projects, but that also explores their links or backlinks.
这些评论可以在整个星期内提炼出来，由每日模板触发，并通过提出随机注释缩小到更小的区域（这得益于重新显示随机页面或块的扩展，如 RoamJS [9] 的 Serendipity）。
These reviews can be distilled throughout the week, triggered by a daily template, and reduced to a smaller area by bringing up a random note (thanks to extensions resurfacing random pages or block, like Serendipity by RoamJS [9]).
评估关系
Evaluate relations
连接数最高的节点排名无疑表明了什么主要调动了我们的精神能量、我们的担忧，甚至我们的痴迷。它是否符合我们的价值观和目的？它是否表明了我所关注的是什么，或者我如何做笔记，或者我用来对信息进行分类的本体论缺乏精确性？
The ranking of the nodes with the highest number of connections certainly gives an indication of what mainly mobilizes our mental energy, our concerns, even our obsessions. Is it consistent with our values and purposes? Is it indicative of what I am concerned with, or how I take notes, or the lack of precision of the ontology I use to classify my information?
仔细观察，哪些节点与边最多的节点相连？最常见的链接是什么？例如，对于页面 [[idea]]，主要链接是什么？[[toRead]] 页面上的链接是否与我当前的目标一致？Mark McElroy 提到他如何记录梦境日记 [10]，并利用 PKG 来利用梦境中出现的符号之间关系的重复性。一点一点地，趋势就会出现，这有助于识别潜意识模式。
Looking more closely, which nodes are connected to the nodes with the most edges? And what are the most frequent links? For example, for the page [[idea]], what are the main links? Are the links on the [[toRead]] page consistent with my current goals? Mark McElroy refers to how he keeps a dream journal [10] and takes advantage of the recurrence of relations between symbols appearing in his dreams thanks to a PKG. Little by little, tendencies appear, which can help to identify unconscious patterns.
对关系的评估也可以侧重于其质量而不是数量。在每周回顾期间，我们可以评估某种关系。它是否对应于分类、查找或表达想法的意图？它表达的是世界实体之间的真实关系还是商品之间的正式关系？它是否对应于一种值得指定并集成到本体中的关系类型？它是否太过单一，以至于无法通过自然语言句子的特定语法来表达？
The evaluation of a relation can also focus on its quality rather than quantity. During our weekly review, we can evaluate a certain relation. Does it correspond to an intention to classify, find or to articulate ideas? Does it express a real relation between entities of the world or a formal relation made by commodity? Does it correspond to a type of relation that deserves to be specified and integrated into an ontology? Is it too singular to be expressed otherwise than through the particular syntax of a natural language sentence?
这些问题很难回答，可能会导致几乎无法解决的哲学问题。如果以这种方式质疑我们图表中的所有关系，那么这项任务将是无法完成的。希望以完全严格、明确和连贯的方式澄清所有关系是不切实际的。澄清的努力仍然是一个开放式的过程，随着新的知识对象的出现和新思想的形成，这个过程不断遇到新的歧义和困难。评估图表中现有关系的工作应该被视为一项不时进行的练习，以巩固一个人构建思想的方式。
These questions are difficult and can lead to almost unsolvable philosophical problems. The task would be insurmountable if it were to question in this way all the relations of our graph. It would be illusory to hope to clarify the totality of the relations in a perfectly rigorous, explicit, and coherent way. The effort of clarification remains an open-ended process, which constantly encounters new ambiguities and difficulties as new objects of knowledge appear and new thoughts are formed. This work of evaluating existing relations in the graph should be seen as an exercise to be done from time to time to consolidate one’s way of structuring one’s ideas.
利用模板和其他自动化功能
Take advantage of templates and other automations
可以创建模板，例如，在模板中询问一些上述关于关系性质的问题，并通过自动查询呈现或提供对目标关系集的快速访问。第 4 章中介绍的智能块可以帮助构建一种方法来查看图的某些部分。然后，我们可以逐步引入或丰富用于记录此信息的模板，从而可能从现有的标准化本体中汲取灵感。[11]
Templates can be created in which, for example, some of the questions mentioned above about the nature of the relations are asked, and which present or offer quick access to a targeted set of relations through automated queries. Smartblocks, which are presented in chapter 4, can help to build a method to review certain parts of the graph. We can then progressively introduce or enrich the templates used to record this information, potentially drawing inspiration from existing and standardized ontologies. [11]
即使我们总是希望并要求做到最好，但我们在很大程度上没有充分利用现有 PKG 系统以及可用模型和本体的潜力。我们经常会固守与更线性的工具相关的习惯。PKG 策划可能需要一些时间才能成为像说话或写作一样普遍的习惯。
We are largely underusing the potential of existing PKG systems and available models and ontologies, even if we always hope and ask for the best. We often remain attached to habits linked to more linear tools. It will probably take time for PKG curation to become a habit as common as talking or writing.
我们可能正处于一种变化的边缘，就像口头文化因书写的发展而发生深刻变化一样。我们太习惯书写了，以至于忘记了书写是一种技术，而书写的日常使用本身就是一种促进技术的发展，它已经彻底改变了我们的思维方式。像所有创新技术一样，书写也遭遇了巨大的阻力。特别是，人们担心它表达了作者不支持或不再支持的思想，就像人们担心 PKG 表达了无意的关系一样。哲学家苏格拉底是最著名的书写批评家之一。我们打算表明，他反对书写的论点不再适用于 PKG 中的书写，或者完全不再适用于 PKG 中的书写，特别是因为无意的关系会引发思考。
We may be on the verge of a change like that which saw oral culture profoundly transformed by the development of writing. We are so used to it we forget that writing is a technology and that its daily use, which itself supposes facilitating technologies, has revolutionized the way we think. Like all innovative technologies, it met with significant resistance. In particular, it was feared it expresses ideas that were not, or are no longer, supported by the author, just as one might fear that a PKG expresses unintended relations. One of the most famous critics of writing is the philosopher Socrates. We propose to show that his arguments against writing no longer apply, or no longer apply entirely, to writing in a PKG, notably because of what unintentional relations bring to thought.
如果苏格拉底能建立一个 PKG，他为什么会写作
Why Socrates would have written if he could have built a PKG
PKG 是思考的合适媒介
PKGs are a suitable medium for thinking
苏格拉底拒绝将他的哲学思考写下来。他总是喜欢口头表达，他认为只有口头表达才能记录下鲜活的思想。如果我们相信柏拉图在《斐德罗篇》[12]中所说的话，他对写作的主要批评是，文本一旦写下来，就不会产生反应。鲜活的思想需要交流，即使不是与真实的人交谈，就像苏格拉底非常喜欢的那样，至少也要进行“灵魂与自己的对话”。[13]
Socrates refused to put his philosophical reflections in writing. He always preferred oral speech, which he claimed is the only one to transcribe the living thought. The main criticism he made of writing, if we believe Plato in the Phaedrus, [12] is that a text, once written, does not react. Living thought requires an exchange, if not with a real person in a conversation, as Socrates liked so much, at least in a “dialogue of the soul with itself”. [13]
写作文本将思想冻结在写作时的状态，赋予其一种坚定和断言的气息，一种陈述真理的姿态，正如苏格拉底在他的对手智者派身上所观察到的那样。一个活生生的思想永远不会停止质疑自己、重新审视自己、澄清自己。
Writing a text freezes the thought in the state it was in at the moment of writing, gives it an air of conviction and assertion, a pretension to state the truth, as Socrates observed in his opponents the Sophists. A living thought never ceases to question itself, to re-examine itself, to clarify itself.
同样，对于读者来说，文本除了一劳永逸地包含的内容外，没有其他要说的。它没有回答问题，没有解释更多，没有做出更多区分，也没有给出更多例子。写作将口头表达的思想的时间动态降低为固定模式，以线性方式组织，固执地从 A 点引向 B 点。
In the same way, for the reader, a text has nothing more to say than what it contains once and for all. It does not answer questions, it does not explain more, it does not make more distinctions, it does not give more examples. Writing reduces the temporal dynamism of the thought expressed in oral speech to a fixed pattern, organized in a linear way, stubbornly leading from a point A to a point B.
当然，人们总是可以回头修改或删除已写的内容。当然，现在有了文字处理器，这比苏格拉底时代更容易，但这仍然只是重写。结果只不过是同一文本的另一个版本，并且具有同样的局限性。
Of course, it is always possible to go back to what one has written and to correct or erase it. That’s easier today with a word processor than it used to be in Socrates’s time of course, but it is still only rewriting. The result is nothing more than another version of the same text and has the same limitations.
在 PKG 中写作提供了一种新的体验。尽管它不能解决所有反对写作的反对意见，并且在响应性和生活交流方面无法与口语相媲美，但 PKG 中的写作不再是一个固定的线性空间。它是一个网络，其节点和关系不断被重塑和丰富。而且这种情况发生的程度比人们想象的要高得多，这是由于半有意和无意关系的不可预见的影响，正如已经表明的那样。
Writing in a PKG offers a new experience. Even though it does not address all objections against writing and cannot compete with oral speech in terms of responsiveness and living exchange, writing in a PKG is no longer a fixed and linear space. It is a network whose nodes and relations are constantly being reshaped and enriched. And this happens to a much higher degree than one would imagine, due to the unforeseen effects of semi- and unintentional relations, as has been shown.
重新考虑一个概念的定义，像苏格拉底经常做的那样，采用二分法，最终可能会得到一个与被判定为误导的概念截然不同的新概念。这会导致所有动员这个概念的判断的含义发生转变。概念的名称甚至可能会改变。这将在图表的所有点上自动更新。
Reconsidering the definition of a concept, proceeding by dichotomy as Socrates often did, can end up with a new conception well distinguished from one that is judged to be misleading. This leads to a shift in the meaning of all the judgments that mobilize this concept. The name of the concept may even change. This will be automatically updated at all the points of the graph.
所有这些判断现在都将与这个新概念以及导致它的讨论直接相关。而一系列已经做出的重新定义本身也可以被记录和浏览：浏览它最终将使我们能够隔离错误的推理并开始一系列新的区分。
All these judgments will now have a direct link with this new concept and the discussion that led to it. And the series of redefinitions that have been made can itself be recorded and browsed: browsing it will eventually make it possible to isolate erroneous reasoning and to start a new series of distinctions.
柏拉图关于苏格拉底的对话可能很适合被铭刻在 PKG 中；它们将成为苏格拉底的一种想象中的 PKG。他每一次长篇离题都试图通过增加问题和区别来定义一个术语，这些离题可以从调动相同概念的另一场对话的任意一点阅读。它们也可以被折回以了解推理的概况，而这种离题常常让我们忘记这一点。
Plato’s dialogues featuring Socrates would probably lend themselves well to their inscription in a PKG; they would be a kind of imaginary PKG of Socrates. Each of his long digressions where he tries to define a term by multiplying questions and distinctions could be read from any point of another dialogue where the same concept is mobilized. They could also be folded back to get an overview of the reasoning, which this digression has often made us forget.
显然，网络上的存在和超文本链接已经允许了这样的路径。PKG 还能给读者带来什么？当然，得益于反向链接、过滤和查询工具，读者可以全方位地阅读。然而，最重要的是，用户可以将部分甚至整个作品整合到自己的 PKG 中。能够将自己的 PKG 与作者的思想相结合，可以为自己的思想带来新的生命和活力。
Obviously, the presence on the Web and the hypertext links already allow such a path. What else does a PKG bring to the reader? Of course, the journey in all directions thanks to backlinks, and filtering and query tools. Above all, however, is the potential for integration of parts or even whole works in the user’s own PKG. Being able to mix one’s own PKG with the author’s thought can bring new life and dynamism to one’s own thought.
写在 PKG 中的想法不再是死的想法。由于自创作以来形成和发展了半有意或无意的联系，它们可以在意想不到的时候重新浮现。上下文的变化可以使它们具有写作时没有的意义。我们可以展开引导我们得出这些想法和论点的线索，并完成或挑战它们。在 PKG 中，与自己的对话因此成为写作过程的同质部分。特别是出于这个原因，我们可以想象苏格拉底不会对此无动于衷，因为他赋予了自我认知决定性的作用。
Thoughts written in a PKG are no longer dead thoughts. They can resurface at unexpected times, due to the semi- or unintentional connections that have formed and developed since their creation. The change of context can make them take on a meaning that they did not have when they were written. We can unfold the thread of ideas and arguments that led us to them, and complete or challenge them. In a PKG, the dialogue with oneself thus becomes consubstantial to the writing process. For this reason in particular, one can imagine that Socrates would not have been indifferent to it, given the decisive role he gave to self-knowledge.
PKG 是了解自己的有效方法
PKGs are a powerful way to know yourself
苏格拉底有一句著名的格言，表达了卓越的哲学追求：“认识你自己！”也就是说，努力思考你所想的，了解你为什么这样想，这对你意味着什么，为什么你重视这个胜过那个。更广泛地说，努力了解作为一个人意味着什么，有弱点也有过头之处。写作，尤其是写日记，长期以来一直是一种强大的工具，可以帮助做到这一点，帮助拉开这种反思所需的距离。
Socrates had a well-known formula, which expressed the philosophical quest par excellence: “Know thyself!” That is, strive to think what you think, to know why you think it, what it means to you, why you value this over that. More generally, strive to know what it means to be a human being, with its weaknesses and its excesses. Writing, and especially journaling, has long been a powerful tool to help do this, to help take the distance necessary for this reflection.
PKG 无疑可以帮助我们走上这条道路。它们让你能够追随你的思路并回到它。此外，PKG 确实记录了你的思想、倾向、偏好、痴迷和思维模式的网络。你思想和你所接受的知识体系中的个人特质不仅体现在内容或表达风格中。它还体现在交织在其中的关系中。一些关系因其重复性而引人注目，而另一些关系则因其不协调性而让你感到惊讶。
PKGs can certainly help us on this path. They allow you to follow the thread of your thought and to come back to it. Moreover, PKGs literally record the network of your thoughts, tendencies, preferences, obsessions and thought patterns. What is personal in your thoughts and the body of knowledge that you embrace is not only in the content or in the style of their expression. It also appears in the relations that are woven into them. Some relations stand out by their recurrence, while others surprise you by their incongruity.
因此，PKG 的图在某种程度上是我们独特性的踪迹，并呈现为元认知的一个显著潜在对象。从这个意义上讲，我们可以非常认真地对待 PKG 中的“P”，因为它不仅涉及我们个人收集的知识，还涉及我们个人和职业生活的各个方面。它还成为自我认知的主要工具之一（也许是最强大的工具）。
Thus, the graph of a PKG is, in a way, a trace of our singularity and presents itself as a remarkable possible object of metacognition. It is in this sense that we can take the “P” in PKG very seriously, in the sense that it is not only about the knowledge that we personally gather, which concerns all aspects of our personal and professional existence. It is also in the sense that it becomes one of the main – and perhaps the most powerful – tools for self-knowledge.
因此，在批评由于过于直观、缺乏条理地使用 PKG 而导致的半意向性关系和无意向性关系的薄弱和泛滥时，我们有点像柏拉图，他接受并发展了苏格拉底对写作的批判，同时他自己也是一位多产的作家，并在 2500 年后仍然是哲学对话写作的典范。有时人们说，柏拉图的哲学创新、他的理想主义和他的方法——辩证法——之所以成为可能，是因为写作，而不是演讲，在他的作品中占有重要地位。例如，沃尔特·J·翁 (Walter J. Ong) 认为：“柏拉图的哲学分析思想，包括他对写作的批判，之所以成为可能，只是因为写作开始对心理过程产生影响。”[14]
Thus, in criticizing the weakness and proliferation of semi- and unintentional relations that would result from the overly intuitive, and not sufficiently methodical, use of PKGs, we would be a bit like Plato taking up and developing Socrates’s criticism against writing while at the same time being a prolific writer himself and remaining a model for the writing of philosophical dialogues 2,500 years later. It is sometimes said that Plato’s philosophical innovations, his idealism, and his method – the dialectic – were made possible by the importance that writing, and not speech, took in his work. For example, Walter J. Ong argues that: “Plato’s philosophically analytic thought, including his critique of writing, was possible only because of the effects that writing was beginning to have on mental processes.” [14]
当我们内化了网络化写作和思考的可能性，当它变得平常时，哪种思想能够通过 PKG 出现？我们才刚刚开始 PKG 的发展。当然，我们的思维方式的变化（如果有的话）将在我们清楚地意识到它们之前很久就发生了。通过有意地在我们的图表中建立链接，我们所做的不仅仅是在其中增加了无意的链接：我们还在我们的大脑中建立了无意的联系，从我们当前的想法到我们未来的想法。我们不仅发展了我们的知识。我们正在开发一个非常适合我们知识发展的环境。通过增加活生生的思想可能采取的路径，我们建立了一个框架，鼓励对我们的知识进行批判性审查，从而将其在强烈的意义上巩固为知识。[15]
Which thought will be able to emerge thanks to PKGs, when we will have internalized the possibilities of networked writing and thinking, and when it will have become ordinary? We are only at the beginning of PKG development. Certainly, however, the changes in our way of thinking, if any, will occur long before we become clearly aware of them. By intentionally establishing links in our graph, we do more than multiply unintentional links in it: we also make unintended connections in our brain, and from our current ideas to our future ideas. We do not only develop our knowledge. We are developing an environment that is highly suitable for the development of our knowledge. By multiplying the paths that living thought may take, we build a framework that encourages the critical examination of our knowledge and thus consolidates it into knowledge in a strong sense. [15]
笔记
Notes
[1] 在本章中，我们所说的 PKG 指的是一组有限的应用，即在过去三年左右变得非常流行的网络笔记系统，例如 Roam Research 或 Obsidian。PKG 还有许多其他形式。我们的讨论重点是用户自己明确地（基本上用自然语言）表述他们记录为个人知识的系统，而不是自动化系统。
[1] By PKG, we refer in this chapter to a limited set of applications, namely the networked note taking systems such as Roam Research or Obsidian that have become quite popular in the last three years or so. There are many other forms of PKG. Our discussion focuses on systems where it is the user themselves who explicitly formulates, essentially in natural language, what they record as personal knowledge, and not an automated system.
[2] 我们使用 Roam Research 特有的语法，通过输入 :: 可以将一个页面作为另一个页面的属性提及。需要注意的是，在 Roam 中，与大多数流行的 PKG 系统一样，关系无法直接确定。需要创建一个间接表达关系的节点。这里的块包含关系类型、“来源”和链接实体（即 URL）的提及。
[2] We use a syntax specific to Roam Research which allows to mention a page as an attribute of another page, by typing ::. It is to be noted that in Roam, as in most popular PKG systems, the relations cannot be determined directly. It is necessary to create a node which indirectly expresses the relation. Here a block contains the mention of the type of relation, “Source” and the linked entity, namely the URL.
[3] https://en.wikipedia.org/wiki/Illusion_of_explanatory_depth
[3] https://en.wikipedia.org/wiki/Illusion_of_explanatory_depth
[4] 这种划分在 PKG 中尤为明显，它同时允许用户以严格的方式构造数据，也为用自然语言表达的想法和链接留下了很大的自由度。Roam Research 中的属性和属性表、Obsidian 中的 Dataview 扩展，或者最近的 Tana 及其超标签继承和类似关系数据库中的结构化表都是 PKG 系统的示例，它们将图的灵活性与更结构化的数据相结合
[4] This division is particularly obvious in PKGs that, at the same time, allow users to structure data in a rigorous way and that also leave great latitude for ideas and links expressed in natural language. Attributes and attributes tables in Roam Research, Dataview extension in Obsidian or, more recently, Tana and its supertags inheritance and structured tables like in a relational database are examples of PKG systems that combine the flexibility of the graph with more structured data
[5] 例如：“我们捕捉和分享的多条信息可以增加偶然联系的频率，尤其是在真正发生创新的组织和学科之间。正如《好点子从何而来》一书的作者史蒂文·约翰逊所说，‘机会青睐有联系的头脑’。” Jarche, H. (2020) Perpetual Beta 2020。https://jarche.com/services/books-in-beta/
[5] For example: “The multiple pieces of information that we capture and share can increase the frequency of serendipitous connections, especially across organizations and disciplines where real innovation happens. As Steven Johnson, author of Where Good Ideas Come From says that, ‘chance favors the connected mind’.” Jarche, H. (2020) Perpetual Beta 2020. https://jarche.com/services/books-in-beta/
[6] 对于古希腊人来说，大自然为每个生命赋予了适当的限度，而人类的愿望往往导致过度，他们称之为傲慢。
[6] In the sense that, for the ancient Greeks, nature gives the limits of what is appropriate to each being, while human aspirations often lead to excess, which they refer to as hubris.
[7] 请注意，Roam Research 可能通过将笔记记录和图形数据库相结合，为 PKG 理念的普及做出了贡献，但它几乎没有开发图形可视化，因为图形可视化本身并没有太大的吸引力。他们用巧妙的非视觉功能弥补了这一不足。其他应用程序，如 Obsidian，在可视化方面做得更好。在本章中，我们决定不将图形可视化视为利用图形中创建的关系的有用方法，因为它们的数量太多，比看起来要复杂得多。然而，我们可以想象，图形可视化管理方面的发展可能会使浏览链接以及直接在图形模式下指定或重新定义它们变得非常有用。
[7] Notice that Roam Research, which probably contributed to popularizing the idea of PKG by combining note taking and a graph database, has hardly developed the visualization of the graph, which, as it is, is without much interest. They have compensated for this lack with ingenious nonvisual features. Other applications, like Obsidian, do much better in terms of visualization. In this chapter, we have decided not to consider the visualization of the graph as a useful way to take advantage of the relations that are created in the graph, since they are far too numerous and more complex than they appear. However, we can imagine that developments in the management of the graph visualization could make it very useful to navigate through the links and to specify or redefine them directly in graphic mode.
[8] https://roamjs.com/extensions/query-builder
[8] https://roamjs.com/extensions/query-builder
[9] https://roamjs.com/extensions/serendipity
[9] https://roamjs.com/extensions/serendipity
[10] McElroy, M. (2022)。如何使用知识图谱（或者为什么知识图谱不仅仅是视觉享受）。
[10] McElroy, M. (2022). How to Use the Knowledge Graph (Or Why The Graph Ain’t Just Eye Candy).
https://markmcelroy.com/how-to-use-the-knowledge-graph-or-why-the-graph-aint-just-eye-candy/
https://markmcelroy.com/how-to-use-the-knowledge-graph-or-why-the-graph-aint-just-eye-candy/
[11] 按照语义网的传统
[11] In the tradition of the semantic web
[12] 柏拉图。《斐德罗篇》275a，《十二卷本柏拉图》（1925 年），第 9 卷，哈罗德·N·福勒译。马萨诸塞州剑桥，哈佛大学出版社；伦敦，威廉·海涅曼有限公司。https://www.perseus.tufts.edu/hopper/text?doc=urn:cts:greekLit:tlg0059.tlg012.perseus-eng1:275
[12] Plato. Phaedrus. 275a, Plato in Twelve Volumes (1925), Vol. 9 translated by Harold N. Fowler. Cambridge, MA, Harvard University Press; London, William Heinemann Ltd. https://www.perseus.tufts.edu/hopper/text?doc=urn:cts:greekLit:tlg0059.tlg012.perseus-eng1:275
[13] “灵魂，正如我所看到的，当它思考时，只是在与自己交谈，向自己提问并回答，肯定和否定。”柏拉图。《泰阿泰德篇》。189e-190a，《十二卷本柏拉图》（1921 年），第 12 卷，哈罗德·N·福勒译。马萨诸塞州剑桥，哈佛大学出版社；伦敦，威廉·海涅曼有限公司。https://www.perseus.tufts.edu/hopper/text?doc=Perseus%3Atext%3A1999.01.0172%3Atext%3DTheaet.%3Asection%3D190a
[13] “the soul, as the image presents itself to me, when it thinks, is merely conversing with itself, asking itself questions and answering, affirming and denying.” Plato. Theatetus. 189e-190a, Plato in Twelve Volumes (1921), Vol. 12 translated by Harold N. Fowler. Cambridge, MA, Harvard University Press; London, William Heinemann Ltd. https://www.perseus.tufts.edu/hopper/text?doc=Perseus%3Atext%3A1999.01.0172%3Atext%3DTheaet.%3Asection%3D190a
[14] Ong, Walter J. (1982). 口语与书写：文字的技术化. 纽约: Routledge.
[14] Ong, Walter J. (1982). Orality and Literacy: The Technologizing of the Word. New York: Routledge.
[15] 我们在第 5 章中解释了强意义上的知识概念。
[15] We explain our conception of knowledge in the strong sense in chapter 5.
第五章
Chapter 5
思维算法，是驾驭 PKG 复杂性的必备 GPS 吗？
Algorithms of thought, the must-have GPS for navigating the complexity of PKGs?
法布里斯·加莱
当我们谈论个人知识图谱 (PKG) 时，我们所说的知识是什么意思？[1] PKG 是我们收集的一组信息，我们认为这些信息如果可供将来使用，将会很有用。这些信息中的许多部分在被收集并至少分类后立即具有知识的价值（例如，URL、推文、博客文章、引言、定义等）。当它们结构清晰并与其他信息相连时，它们会获得更大的价值。众所周知，这些联系可以促进理解、记忆和未来使用。这当然是 PKG 界面所促进和支持的。当需要思考或行动时，PKG 有望简化信息的重新浮现。理想的情况是，当你不知道需要它时，它们会重新浮现相关信息，从而有助于原创见解或创新。
What do we mean by knowledge when we talk about Personal Knowledge Graphs (PKG)? [1] A PKG is a set of information we have collected that we think will be useful if it is available for future use. Many of these pieces of information immediately have the value of knowledge as soon as they are collected and at least categorized (e.g., an URL, a tweet, a blog post, a quote, a definition, etc.). They gain even more value when they are clearly structured and connected to other information. These connections, as is generally acknowledged, can facilitate understanding, memorization, and future use. This is of course what the PKG interface facilitates and supports. PKGs are expected to ease the resurfacing of information when it is needed to think or act. And the ideal is that they resurface relevant information when you did not know it was needed, contributing to original insight or innovation.
除了将知识视为在适当情况下可用的信息这一概念之外，我们还想探索 PKG 如何在知识的构成中发挥作用，这种知识在更深层次的意义上被理解为对世界的有理有据的判断，或者用柏拉图的定义来说，是“合理的真实信念”[2]（见图 5.1）。寻找坚实而严谨的论证是科学的关注点。但在反思 PKG 的背景下，日常生活中的所有观点和所有有争议的判断都可以被视为构成个人知识。只要我们准备好认领它们，它们就成为我们个人知识的一部分，它们决定了我们的思维和生活方式。
Beyond this conception of knowledge as information that is made available in the appropriate context, we want to explore how PKGs can play a role in the constitution of knowledge understood in a stronger sense, that of a well-founded judgment about the world or, to use Plato’s definition, “a justified true belief” [2] (see figure 5.1). The search for a solid and rigorous justification is the concern of science. But within the context of a reflection on PKGs, all opinions and all disputable judgments from everyday life can be considered as constituting personal knowledge. They are a part of our personal knowledge as soon as we are ready to claim them, and they determine the way we think and live.
图 5.1 知识作为真实的、经过证实的信念
Figure 5.1 Knowledge as True Justified Belief
PKG 如何有助于我们知识的构建，并帮助我们挑战和建立其健全性？我们将尝试表明，借助模板，更广泛地说，通过调动思维算法 (AoT)，我们可以引导我们的思维朝着有利于构建更健全知识的步骤迈进。
How can PKGs contribute to the constitution of our knowledge, and help us in challenging and building its soundness? We will try to show that with the help of templates and more generally by mobilizing algorithms of thought (AoT) we can guide our thinking towards steps that favor the constitution of more sound knowledge.
PKG 中的 K 具有强烈的意义：不仅仅是结构化数据
K in PKG in the strong sense: More than structured data
二手知识和传统知识
Secondhand and conventional knowledge
PKG 可以在一张图中收集其创建者生活的各个方面的信息，它可能包含各种类型的信息。其中大部分是我们在网络上漫游时收集的信息：URL、引文、推文、阅读笔记、收听播客、观看视频、参加在线课程、摘要等。如果可以将此类信息视为知识，那么它主要是二手知识。通常，这种类型的信息是知识的弱化形式：它通常是指知识来源，是指向可用知识的标志，但我们不能说我们拥有这些知识。我们可以说我们知道在需要时在哪里找到它，但这并不能保证我们能够理解它并利用它。如果我们了解了作者对世界的看法，我们知道的是他们声称了这一点，但这并不等同于对世界的直接了解，因为我们不能立即确定作者是否掌握了关于世界的真理，我们是否正确理解了他们所说的话。我们可以说我们知道他们说了什么，但我们仍然需要对其进行检验，最终将他们的判断整合到我们对世界的认识中。[3]
A PKG can gather information on all aspects of the life of its creator in a single graph and it may contain a wide variety of types of information. Much of it is information collected throughout our wanderings on the web: URLs, quotes, tweets, notes from reading, listening to podcasts, watching videos, taking an online course, summaries, etc. If such information can be considered as knowledge, it is primarily in the sense of secondhand knowledge. Often, this type of information is knowledge in a very weak sense: it generally refers to a source of knowledge, it is a sign that points to available knowledge, but we cannot say that we possess this knowledge. We can say we know where to find it when needed, which does not guarantee that we will be able to understand it and take advantage of it. If we have learned what an author says about the world, what we know is that they claim it, but that does not equate to direct knowledge about the world, since we cannot be immediately sure that the author has grasped a truth about the world and that we have correctly understood what they say. We can say that we know what they say, but we still need to examine it to eventually integrate their judgments into our knowledge about the world. [3]
有一种非常常见的情况，即二手知识和直接知识在未经证实的情况下是等同的：即传统真理的情况。例如，如果一种编程语言的参考页面说某某命令允许某某语法或参数，则该来源仅说明了使用该命令必须遵循的惯例。传统真理是一种不成问题的知识，在 EKG 和专注于传统主题（例如外语或编程语言、法律文本、说明书等）的人的 PKG 中很突出。
There is a very common case where secondhand knowledge and direct knowledge, without justification, are equivalent: the case of conventional truths. For example, if a reference page of a programming language says that such and such a command allows such and such syntax or arguments, this source only states the convention that one must follow to use this command. Conventional truths are a type of unproblematic knowledge that is prominent in EKGs and in the PKG of people focused on conventional topics such as a foreign or a programming language, a law text, an instruction manual, etc.
直接和二手知识不需要证明。它们的价值在于我们将如何利用它们。我们需要做的就是记住它们或在合适的时间找到它们。为了便于访问和利用，它们必须结构良好。这可能意味着对它们进行分类并突出显示最重要的信息。它还意味着将它们与其他相关信息联系起来，以提供多种访问路径并提高搜索引擎查询的可查找性。
Direct and secondhand knowledge do not require justification. Their value is in the use we will make of them. All we need to do is to memorize them or find them at the right time. To be accessible and exploitable, it is important they are well structured. That can mean categorizing them and highlighting the most important information. It can also mean connecting them to other relevant information to offer several access paths and improve findability for search engine queries.
需要论证的知识
Knowledge requiring justification
关于世界的判断如果没有论证，就不能严格地被视为知识。例如，如果我们想把它们视为知识，那么像“冠状病毒主要通过空气传播”这样的事实陈述或像“你应该在运动后伸展一下以更快地恢复”这样的行动原则就不能简单地被接受。如果没有论证（即使它们恰好是真的），它们仍然是一种简单的信念或观点。
Judgments about the world cannot be rigorously considered as knowledge without justification. For example, a factual statement like “the coronavirus is mainly transmitted through the air” or a principle of action like “you should stretch after sport to recover more quickly” cannot be simply admitted if we want to consider them as knowledge. They remain a simple belief or opinion in the absence of justification (even if, by coincidence, they are true).
我们所谓的知识，在强意义上，是合理的判断，即我们认为相当有根据的判断，或者至少在我们看来，根据现有信息，这些判断比其他判断更有根据。从这个意义上讲，知识需要思想运动来奠定基础，同时也需要比较、质疑并最终批评它。为了引导我们的思维并在把握现实的方式上取得进步，我们必须意识到我们倾向于支持的主要主张以及我们必须进一步研究的信念。重要的是，这组知识通过专用标签或页面在我们的 PKG 包含的全部信息中脱颖而出。
What we call our knowledge, in the strong sense, are sound judgments, i.e., judgments that we consider to be fairly well founded, or at least that seem to us to be better founded than competing judgments, given the information available. In this sense, knowledge requires a movement of thought to ground it but also to compare it, question it and eventually criticize it. To orient us in our thinking and to make progress in the way we grasp reality, we must be aware of the main claims that we are inclined to support and of the beliefs we must examine further. It is important that this set of knowledge stands out within the whole of the information that our PKG contains by means of dedicated tags or pages.
需要论证的知识可以有多种形式。为简单起见，我们将主要关注主张，偶尔关注证据和概念。主张可以是有关世界的特定陈述（例如，“我有乳糖不耐症”）或一般陈述（例如，“加糖食物并不比高碳水化合物食物更健康”）。它们也可以是抽象原则（例如，“任何血糖峰值之后都会出现成比例的低血糖阶段”）、定律、理论或实用规则（例如，“以生蔬菜开始进餐以降低血糖峰值”）。证据是观察到的事实。它们需要得到论证，因为我们的感觉并不完全可靠，描述它们的话语可能会产生误导。概念是一种思考给定类型对象的方式，一种定义它并将其与其他对象区分开来的方式。它们需要得到论证，因为定义（当它不是常规定义时）是一种主张：它是关于给定类型对象的核心现实的主张。使用给定的概念就是论证它比其他概念更符合现实。我们可以补充说，问题是对矛盾的认识或我们知识的意外限制，但这仍然是一种主张，一种元认知的主张。
Knowledge requiring justification can take many forms. For simplicity, we will focus mainly on claims, and occasionally on evidence and concepts. Claims can be a particular statement (e.g., “I am lactose intolerant”) or a general one (e.g., “sweetened foods are not healthier than high-carbohydrate foods”) about the world. They can also be abstract principles (e.g., “any peak in blood glucose is followed by a proportional phase of hypoglycemia”), laws, theories, or practical rules (e.g., “start meals with raw vegetables to reduce the blood sugar peak”). Evidence is facts that have been observed. They need to be justified because our senses are not entirely reliable, and the discourse that describes them can be misleading. Concepts are a way of thinking about a given type of object, of defining it and distinguishing it from others. They need to be justified since a definition, when it is not conventional, is a kind of claim: it is a claim about the core reality of the given type of object. To use a given concept is to argue that it fits better with reality than other concepts. We could add that a problem is the knowledge of a contradiction or unexpected limits of our knowledge, but this is still a kind of claim, a metacognitive one.
总而言之，我们的个人知识由一系列主张、我们视为证据的数据以及我们用来表达这些主张的概念组成，只要我们认为这些主张有充分的依据。这组陈述构成了 PKG 的核心，它不仅是二手信息的总和，而且是我们采纳的、可以断言为对现实的良好描述的想法。所有这些元素都可以对应于科学知识，以它们为指导总是好的。然而，既然我们谈论的是 PKG，那么就不是一个遵守学术知识要求的问题。我们谈论的是与我们的日常生活、我们的判断、我们在健康、人际关系、政治等所有生活领域的决定直接相关的信念。因此，这些是我们普遍支持的主张，但不一定以严格的证据为基础。
To summarize, our personal knowledge is made up of the set of claims, the data that we take for evidence, and the concepts we use to express them, insofar as we consider them to be sufficiently well founded. This set of statements constitutes the heart of a PKG, which is not only the sum of secondhand information but the ideas we have adopted and can assert as a good description of reality. All these elements can correspond to scientific knowledge and it is always good to be guided by them. However, since we are talking about a PKG, it is not a question of complying with the requirements of academic knowledge. We are talking about beliefs that directly concern our daily life, our judgments, our decisions in all areas of life, such as health, relationships, politics, etc. These are therefore claims that we commonly support without necessarily grounding them in rigorous evidence.
我们可以进一步区分与我们的知识相关的两个重要子集，它们构成了两个层次：使用 Reichenbach 引入的词汇，[4] 我们不能混淆发现的背景（主要与来源有关）和论证的背景（重建判断逻辑的努力）。PKG 有助于记录这些不同的背景，这要归功于它们使我们能够创建和浏览的丰富而交织的结构。通过努力确定我们的哪些知识最有根据，并养成更好地论证它的习惯，我们将提出一套方法来创建有利于健全知识的论证背景。这与构成 PKG 大部分的更混乱和不可预测的信息和想法的积累不同。我们将重点关注模板，尤其是当它们可以发挥 AoT 的作用时。
We can further distinguish two important subsets related to our knowledge, which constitute two layers: to use the vocabulary introduced by Reichenbach, [4] we must not confuse the context of discovery (mainly related to the sources) and the context of justification (the effort to reconstruct the logic underlying a judgment). PKGs facilitate the recording of these distinct contexts, thanks to the rich and intertwined structures that they enable us to create and to go through. By working to identify which of our knowledge is best grounded and developing habits to better justify it in general, we will propose a set of ways to create a context of justification that is favorable to sound knowledge. This is distinct from the more disordered and unpredictable accumulation of information and ideas that make up the greater part of a PKG. We are going to focus on templates, especially when they can play the role of AoT.
什么是思维算法？为什么它们在 PKG 环境中有用？
What are the algorithms of thought and why are they useful in the context of PKGs?
模板作为图层
Templates as graph layer
模板使我们能够调用预定义的结构，通常被视为文档模型或数据库中表字段的等价物。从 PKG 的角度来看，它更像是提供一个层，该层预定义了与图中某些节点的一组关系。它们当然有助于以同质的方式构建某些类型的知识，特别是我们指定为二手或传统知识的知识。例如，在文章页面顶部记录一系列属性很方便，例如完整标题、作者、出版日期、主题、阅读状态等。该元数据可以通过笔记记录的指导方针来完成，例如邀请写一个简短的摘要、三个最重要的想法、一个或几个有疑问的想法、学到了什么等。由于模板的每个元素都可以预先链接到图的某些节点，因此笔记记录已经注册在图的多个部分中。
Templates enable us to invoke a predefined structure, often seen as a document model or as the equivalent of table fields in a database. From the perspective of PKGs, it is more like providing a layer that predefines a set of relationships with certain nodes in the graph. They are of course useful for structuring certain types of knowledge in a homogeneous manner, in particular that which we have designated as secondhand or conventional knowledge. For example, it is convenient to note, at the top of a page on an article, a list of properties like the complete title, the author, the date of publication, the theme(s) addressed, the reading status, etc. That metadata can be completed by the guidelines of the note-taking, for example an invitation to write a brief summary, the three most important ideas, one or a few questionable ideas, what was learned, etc. As each element of the template can be in advance linked to some nodes of the graph, the note-taking is already registered in a multitude of parts of the graph.
模板和创造力
Templates and creativity
模板的众所周知的缺点是它需要使用预定义的结构，这会将注意力引导到一组不一定最适合手头特定对象的方向。当对象类型复杂、不为人所知或随着时间的推移而被发现时，情况尤其如此。这会降低适应性并影响创造力。PKG 的灵活性使得我们能够通过轻松重新配置模板提出的连接来适应逐渐出现的结构来克服这个问题，但模板对思维的限制也是不可忽视的。
The well-known disadvantage of a template is that it requires the use of a predefined structure, which directs the attention in a set of directions that is not necessarily the most adapted to the particular object at hand. This is especially true when the type of object is complex, not well known or is discovered over time. This can reduce adaptability and impact on creativity. The flexibility of the PKGs makes it possible to overcome this problem by easily reconfiguring the connections proposed by a template to adapt to the structure that gradually emerges, but the constraint on thinking caused by a template is not negligible.
思维受制于某种惯性。如果没有东西来挑战或撼动它，它就会一直依附于其主导的思维模式。我们可能担心模板会强化这种倾向。相反，如果没有模板提供启动写作或思考的提示，我们可能担心思维会保持被动，例如，只会收集引语。因此，最好设计鼓励思考而不是限制思考的模板。一个简单的技巧是提出开放式问题，而不是给出处方。就像诗歌写作的限制不会阻碍创造力，相反，它为玩弄语言提供支持一样，模板的提示应该激发思想，鼓励它采取流动性，使其尽可能贴近其对象；简而言之，敢于思考，而不是让自己被动地被引导。这正是良好的思想指导方针可以做出的贡献，正如下一节所讨论的那样。
The mind is subject to a kind of inertia. It remains attached to its dominant models of thought if nothing challenges it or shakes it up. We may fear that a template will reinforce this tendency. Conversely, without a template that offers prompts to initiate writing or thinking, we may fear that the mind will remain passive and merely collect quotes, for example. It is therefore better to design templates that encourage thinking without limiting it. A simple technique is to ask open-ended questions rather than giving a prescription. In the same way that the constraints of poetic writing do not prevent creativity but, on the contrary, provide support for playing with language, the prompts of a template should motivate thought, incite it to adopt the fluidity that will allow it to stick as closely as possible to its object; in short, to dare to think rather than to let itself be passively guided. This is precisely what good guidelines for thought can contribute to, as discussed in the next section.
思维算法作为思维的动态模板
Algorithms of thought as dynamic templates for the mind
除了数据结构之外，知识构成中似乎最基本的是建立或讨论该知识的一组智力操作。这些操作必须遵循一定的顺序才能获得一定的结果。这种顺序被系统化为一系列有限的定义步骤，在此期间，注意力集中在某个对象上并执行智力操作，这对应于所谓的思维算法 (AoT)。[5] 它们本质上是支持有序思维模式的模板。它们是有效执行某种智力活动的方法的模型。我们将在下一节中介绍一些示例。
Beyond a data structure, what seems to be primordial for the constitution of knowledge is the set of intellectual operations that establish or discuss that knowledge. These operations have to follow a certain order to obtain a certain result. Such an order, schematized in a finite series of defined steps, during which the attention is directed to a certain object and an intellectual operation is performed, corresponds to what has been called an Algorithm of Thought (AoT). [5] They are, in essence, templates to support an ordered thought pattern. They are the modeling of a method to efficiently perform a certain type of intellectual activity. We will present some examples in the following section.
训练我们的推理能力，以便尽可能明智地做出判断并不容易。尽管我们提出的每个操作或问题可能看起来简单或微不足道，但我们通常不会花精力去检验我们的主要观点。然而，这是必不可少的。
Disciplining our reasoning to make judgments with as much discernment as possible is not easy. Although each operation or question we ask may seem simple or trivial, the effort required to examine our main opinions is not so commonly provided. It is however essential.
能够自发地一次又一次地将分散的注意力拉回来的能力，是判断力、性格和意志的根源。如果没有这种能力，任何人都不会是有能力的。能够提高这种能力的教育才是卓越的教育。但定义这一理想比给出实现这一理想的具体指导要容易得多。[6]
The faculty of voluntarily bringing back a wandering attention, over and over again, is the very root of judgment, character, and will. No one is compos sui [competent] if he [doesn’t have it]. An education which should improve this faculty would be the education par excellence. But it is easier to define this ideal than to give practical directions for bringing it about. [6]
如果 AoT 经过深思熟虑并很好地融入 PKG，它们可以为这样的教育做出贡献。所提出的模板——可能看起来僵化、简单，或者相反，过于沉重，无法应用于每一条知识——应该被视为一种练习，让我们学会将注意力集中在一系列心理操作上，从而加强大脑在 PKG 内外更牢固地积累知识的倾向。正如我们对模板所说的那样，对于 AoT 来说更是如此，它是一种推动，一种邀请，而不是一个需要机械遵循的僵化框架——这将产生与我们寻求的效果相反的效果，即启动一场真正的思想运动。
AoTs could contribute to an education like this if they are well thought out and well integrated into a PKG. The proposed templates – which may seem rigid, simplistic, or, on the contrary, excessively heavy to apply to each piece of knowledge – should be considered as exercises to learn to maintain our attention on a sequence of mental operations and thus reinforce the mind’s disposition to build up its knowledge more solidly, in the PKG as well as outside. As we said for templates, and it is even more true for an AoT, it is a nudge, an invitation, more than a rigid framework to be followed mechanically – which would have the opposite effect to the one we are looking for, namely initiating an authentic movement of thought.
为什么 AoT 在 PKG 环境中很有用
Why AoTs are useful in the context of PKGs
在 PKG 中，知识以非线性的方式表达，找到方向并不总是那么容易。传统教育并没有让我们习惯这种结构。因此，我们需要指南——“地标”。通往图中其他节点的大量链接也很容易分散注意力：可能有上千条路径。有时迷路是可以的，但更常见的情况是，你必须知道如何定位自己。正如我们上面看到的，模板层可以帮助实现这一点：它突出显示指向图中某些节点的链接选择，从而引导用户的注意力。如果我们需要一种方法来穿越复杂的道路，那么必须在不复制固定、封闭的框架的情况下完成。这正是 PKG 允许我们做的事情。因此，动态、模块化的模板是可取的。最后，由于 PKG 的开发首先是一个个人过程，需要你长期投入其中，因此它不应该是一种非个人化的记录：思考必须无处不在，目标确实是更好地思考，更好地发展你的思维，而这正是 AoT 可以帮助实现的。在下一章中，我们将介绍几个 AoT，它们专门用于指导我们认为在知识构成中很重要的要点的思考。比预先建立的结构更重要的是邀请人们努力思考，特别是建立与图表其他元素的链接。
In a PKG, the knowledge is articulated in a nonlinear way, it is not always easy to find your way around. Classical education does not accustom us to this type of structure. We therefore need guides – “landmarks”. The multitude of links available to other nodes of the graph can also easily disperse attention: a thousand paths are possible. It is sometimes appropriate to get lost, but more usually, you must know how to orient yourself. That can be helped by the template layer, as we have seen above: it highlights a selection of links to certain nodes of the graph and thus guides the user’s attention. If we need a means to find our way through the complexity, it must be done without reproducing a fixed, closed framework. This is precisely what the PKGs allow us to do. Dynamic, modular templates are therefore desirable. Finally, since the development of a PKG is above all a personal process, to which you commit yourself durably, it should not be an impersonal recording: thinking must be present everywhere, the goal is indeed to think better, to better develop your thinking, which is what AoTs can help to achieve. In the following chapter, we will present several AoTs that are specifically intended to guide thinking on points we consider important in the constitution of our knowledge. What counts more than the preestablished structure is the invitation to make an effort to think and in particular to establish links to other elements of the graph.
用于构建知识的 AoT 示例
Examples of AoTs to build knowledge
当某种类型的新知识被记录到数据库中时，通常会输入一组属性。这使我们能够以同质的方式收集知识所需的主要信息，以便知识足够明确并在未来快速利用。在这个阶段，经典数据库的表格或 PKG 中模板调用的属性列表（如下例所示）并没有根本区别。经典数据库中的表格可以提供更清晰的框架，而在关系数据库中，很容易从链接表中调出有用的信息。这在大多数 PKG 中都不太简单，尽管有间接的方法或扩展可以做到这一点。[7] PKG 的优势当然是灵活性：您可以自由添加或删除属性、重新排列它们、在每个属性下插入注释等。任何元素都可以链接到任何其他元素或多个其他元素，具体取决于每个对象的需求和特殊性。浏览这些链接是直观的。这种灵活性更好地满足了知识的要求，因为现实总是比我们试图将其挤入的盒子更丰富、更多样化。但另一方面，由于缺乏相同且可靠的结构，因此存在一定的复杂性和缺乏参考点。逐行阅读表格中的行已经不够了。这就是为什么值得使用 AoT 的原因。
When new knowledge of a certain type is recorded in a database, a set of properties is generally entered. This enables us to gather in a homogeneous way the main information required for the knowledge to be explicit enough and quickly exploitable in the future. At this stage, the table of a classic database or a list of properties called up by a template in a PKG, as in the examples below, do not differ fundamentally. The table in a classic database can provide a clearer framework, and in a relational database it is easy to bring up useful information from a linked table. This is less straightforward in most PKGs, although there are indirect ways or extensions to do this. [7] The advantage of PKGs is of course the flexibility: you can freely add or remove properties, rearrange them, insert comments under each property, etc. Any element can be linked to any other or several others, depending on the needs and particularity of each object. And browsing these links is intuitive. This flexibility better meets the requirements of knowledge, because reality is always richer and more varied than the boxes we try to squeeze it into. The other side of the coin is a certain complexity and a lack of reference points due to the lack of identical and reassuring structures. It is no longer enough to read the lines of a table one after the other. This is why it is worthwhile to use AoTs.
以下所有示例模板均在 Roam Research 中设计。[8] 其中大多数将采用 SmartBlocks [9] 语法。 我们的目的不是详细解释此语法，而是评论此处使用的主要功能。
All the following example templates have been designed in Roam Research. [8] Most of them will adopt the SmartBlocks [9] syntax. The goal is not to explain this syntax in detail, but we will comment on the main features used here.
不同类型知识的模板示例
Example of templates for different types of knowledge
以下是记录思想实验的模板示例。它通常是二手知识（除非你自己发明的）。
Here is an example of a template for recording a thought experiment. It is usually (unless you invented it yourself) secondhand knowledge.
图 5.2 思想实验的简单模板
Figure 5.2 Simple template for Thought experiments
它是一个模板，应该插入到页面中，页面的标题是思想实验的简短描述。由于一些次要的原因，具体到 Roam 后端以及数据库查询或过滤器的工作方式，我们将所有属性缩进了一个主要属性下，该属性描述了相关对象的类型。每个属性（在 Roam 中称为“属性”）都是可点击的，并允许用户查看使用它们的所有页面的列表。因此，相关标签（或页面，因为它们只是 Roam 中指代同一事物的两种不同语法）可以在很大程度上是横向的。它们可以描述该字段，以及任何元素，这将增加此页面在有用时重新出现的机会。
It is a template that should be inserted in a page whose title is a short description of the thought experiment. For some reasons that are secondary here, specific to the Roam backend and the way the database queries or filters work, we have indented all the properties under a main property that characterizes the type of object in question. Each property (named “attribute” in Roam) is clickable and allows users to see the list of all pages that use them. Related tags (or pages, since they are just two different syntaxes that refer to the same thing in Roam) can thus be largely transversal. They can describe the field, as well as any element that will increase the chances of having this page resurfaced when it could be useful.
该模板的每个块都构成图形的一个或多个节点。例如，“作者”块显然允许我们将“作者”属性链接到给定的作者实例，并将该作者链接到这个思想实验。每个节点都是图形另一部分的入口点，可以轻松浏览，而无需设计复杂的查询或两个数据表之间的连接。例如，所有与作者相关的数据都通过“作者”节点链接到这个思想实验。因此，该模板构成的简单且易于模块化的表示涵盖了丰富的图形结构，我们在图 5.3 中表示了其中的一小部分。
Each block of this template constitutes one or several nodes of the graph. For example, the block “Author” allows us to both, obviously, link the property “Author” to a given author instance and link that author to this thought experiment. Each node is an entry point to another part of the graph, which can be browsed easily without designing a complicated query or a join between two data tables. For example, all the data related to the author are linked, via the “Author” node, to this thought experiment. The simple and easily modulable presentation this template constitutes thus covers a rich graph structure, of which we represent a small part in the figure 5.3.
图5.3 模板下方的部分图形结构
Figure 5.3 Part of the graph structure below the template
以下是最常见的需要论证的个人知识类型的模板：主张。[10] 当然，人们可以简单地将关键字“主张”添加到陈述主张的区块中，并将主张中的某些单词转换为页码引用，这将有助于重新出现它们。但是，使用以下模板可以为将来的评估、论证和更好地响应查询奠定基础：
Here is a template for the most common type of personal knowledge that needs some justification: a claim. [10] One could, of course, simply add the keyword “claim” to the block in which the claim is stated and turn some of the words in the claim into a page reference, which will facilitate their resurfacing. With the following template, however, the basis is set for future evaluation, justification, and better responsiveness to queries:
图 5.4 索赔的简单模板
Figure 5.4 Simple template for Claims
“健全性”属性可以精确评估该主张的可靠性。与之关联的标签 [11] 表示，对于最终查询，该主张仍有待评估。“什么支持这个说法？”是一个开放式问题，鼓励我们思考理由。
The “Soundness” attribute enables a precise evaluation of how solidly the claim is founded. The tag [11] associated with it indicates, for eventual queries, that the claim remains to be evaluated. “What supports this statement?” is an open question that encourages us to think about justifications.
前面的例子是经典的静态模板。属性的顺序（明确或隐含地对应于一系列问题）已经类似于 AoT。但我们接下来要介绍的例子是动态模板，可以更紧密地引导思维过程。
The previous examples are classic, static templates. The order of the properties, which corresponds explicitly or implicitly to a series of questions, already resembles an AoT. But the examples that we will present in the following are dynamic templates, which can steer the thought process more closely.
用于记录一般知识的 AoT 示例
Example of an AoT to record knowledge in general
在完成模板之前，记录知识所需的第一个思考工作是识别我们要记录的对象类型，并打开相应的模板，而这可以实现自动化。最重要的是，介绍这个 AoT 是一个提供条件算法示例的机会，因为这个操作仍然很简单，我们可以轻松地在不自动化的情况下完成它。
The first effort of thought necessary to record knowledge, before completing a template, and which can be automated, is the identification of the type of object that we are about to record, and the opening of the corresponding template. Presenting this AoT is, most importantly, the opportunity to give an example of a conditional algorithm, because this operation remains elementary, and we can easily do without automating it.
假设我们有几种类型的知识对象，包括主张和思想实验，以及相应的模板。我们想要为某种类型的知识对象创建一个新页面，并在该页面上自动插入相应的模板。为此，我们希望获得已定义模板的知识对象列表（否则，我们将在创建新模板之前单击取消）。选择对象类型后，系统将要求我们在图形中输入新对象的名称（即与其对应的页面的名称）。然后将创建一个新页面，将相应的模板复制到该页面，并自动在右侧边栏中打开该页面（并将启动此模板的自动操作，我们将在后面看到）。
Let’s say we have several types of knowledge objects, including claims and thought experiments, and the corresponding templates. We want to create a new page for a certain type of knowledge object, on which the appropriate template will be automatically inserted. To do this, we would like to be offered the list of knowledge objects for which we have defined a template (otherwise, we will click on cancel, before creating a new template). After choosing the type of object, we will be asked to enter the name of the new object in the graph (i.e., the name of the page that will correspond to it). Then a new page will be created, the corresponding template will be copied to it, and the page will automatically be opened in the right sidebar (and the automatic operations of this template will be launched, as we will see later).
图 5.5 带条件的 SmartBlocks 界面和结果
Figure 5.5 Interface and result of the SmartBlocks with condition
图 5.6. 带条件的智能块
Figure 5.6. SmartBlocks with condition
这里使用的语法是 SmartBlocks 的语法。是的，对于非程序员来说，这可能会令人害怕！我们可能希望语法更简单，但通过一些练习和一些示例，每个人都可以理解。这些命令通常是不言自明的，并且对应于基本操作（例如请求用户输入），或对应于图表中可用数据的占位符，例如另一个模板或当前日期。非技术用户每天都会在电子表格公式中使用类似的语法，或者配置在 Readwise 中读取的文本的元数据。无论如何，理解此语法对于进一步阅读来说不是必需的，您可以跳过以下详细介绍的段落。
The syntax used here is that of SmartBlocks. Yes, it can be scary for noncoders! We may wish the syntax was simpler, but with a little practice and some examples, it’s quite understandable for everybody. The commands are usually self-explanatory and correspond either to basic operations such as requesting input from the user, or to placeholders for data that is otherwise available in the graph, such as another template or the current date. Nontechnical users use similar syntax on a daily basis in spreadsheet formulas or to configure the metadata of texts read in Readwise, for example. Anyway, understanding this syntax is not necessary to read further, you can skip the following paragraphs that go into a bit of detail.
该代码分为以下几个元素：
The code breaks down into the following elements:
<%SET: : 设置变量的值，该变量的名称在 : 后指定，值在 , 后指定。这里我们依次定义用户的选择和要创建的对象的名称
<%SET: : sets the value of a variable whose name is specified after : and the value after the ,. Here we define successively the choice of the user and the name of the object to create
<%INPUT:: 打开一个对话框，在 : 后显示问题。用户需要从以 %% 分隔的元素列表中选择对象类型。然后输入要创建的对象的名称。
<%INPUT:: opens a dialog box with the question indicated just after :. The user is invited to choose the type of object from a list of elements separated by %%. Then they type name of the object to create.
<%NOBLOCKOUTPUT%> 只是表示此行不会导致图中创建任何块，因为它用于定义此 AoT 的变量。
<%NOBLOCKOUTPUT%> simply indicates that this line will not result in any block creation in the graph, since it is used to define the variables of this AoT.
一系列 <%IFVAR:variable,value%>：此块将以条件方式考虑：如果指示的“变量”具有指示的“值”，则将其考虑在内，如果没有，则转到下一个。具体来说，如果用户选择了一种对象类型，则 sb 变量（用于 SmartBlock）将具有该值。
A series of <%IFVAR:variable,value%>: this block will be taken into account in a conditional way: if the indicated “variable” has the indicated “value”, then it is taken into account, if not, we go to the next one. Concretely, if the user has chosen a type of object, then the sb variable (for SmartBlock) will have that value.
<%GET:variable%> 检索先前由 <%SET%> 定义的变量的值并将其插入到块中的指定位置。
<%GET:variable%> retrieves the value of the variable previously defined by <%SET%> and inserts it into the block at the specified location.
<%TAG:foo%> 插入“foo”作为参考页面：[[foo]]。（这避免创建与模板相关的链接。）
<%TAG:foo%> inserts “foo” as a reference page: [[foo]]. (This avoids creating a related link to the template.)
<%SMARTBLOCK:workflow,page%> 在指定页面上运行指定的工作流程。
<%SMARTBLOCK:workflow,page%> runs the specified workflow on the indicated page.
<%SIDEBARWINDOWOPEN:page%> 打开侧边栏中指定的页面。
<%SIDEBARWINDOWOPEN:page%> opens the page specified in the sidebar.
请注意，有一个条件结构：如果知识对象属于给定类型，则运行相应的模板。通过这种逻辑，我们可以构建相当复杂的 AoT，其中考虑了大量嵌套选项。一个选择或一系列条件选择允许思想沿着图表的不同路径移动，从而提供上下文中的相关信息。例如，在这里您可以轻松查看其他声明是否具有相同的“相关问题或难题”，或者关注其中一个“相关标签”以查看是否已将有用信息添加到图表中。遵循这种条件模板不会将思想锁定在固定的结构中，因为一旦显示结构，就总是可以修改结构，并根据经验更改或改进模板。事实上，图表允许在不丢失信息的情况下重新建模数据，并且独立于其以前的结构。
Notice that there is a conditional structure: if the knowledge object is of a given type, then run the corresponding template. With this logic, we can build quite complex AoTs, which take into account a multitude of nested options. A choice or a series of conditional choices allows the thought to move down different paths of the graph, which makes available the relevant information in the context. For example, here you can easily see if other claims have the same “Related questions or problems” or follow one of the “Related tags” to see if useful information has already been added to the graph. And following this kind of conditional template doesn’t lock the thought into a fixed structure, since it is always possible to modify the structure once it has been displayed, and to change or improve the template with experience. Indeed, a graph allows the remodeling of data without loss of information and independently of its former structure.
在稍微复杂一点的版本中，我们添加了“其他”对象类型选项，提示用户从通用模板创建新模板：
In a slightly more complicated version, we have added an “Other” object type option, which prompts the user to create a new template from a generic template:
图 5.7 SmartBlocks 构建另一个 SmartBlocks
Figure 5.7 SmartBlocks building another SmartBlocks
这是创建特定类型模板的模板示例！这允许尊重预定义的本体（可能更复杂，有一系列选择）并保持一定的同质性，但这个框架永远不会固定，因为它很容易修改。
This is the example of a template to create a certain type of template! This allows to respect a predefined ontology (which could be more complex, with a series of choices) and to keep a certain homogeneity, but without this framework ever being fixed since it is easily modifiable.
促使人们首先考虑最重要的事情的 AoT 示例
Example of AoT that prompts one to consider the most important first
一旦我们确定了对象的类型并插入了模板，我们就需要完成它。在我们的思想实验示例中，关键点是相对标签（或概念）和相对问题或难题 [12]，这将使我们能够在以后找到它们，尤其是当我们没有寻找它但它可能有用时。完成这些字段也是需要最大智力努力的事情。为了方便或缺乏时间，这些字段可能会留空或在开始时匆忙完成。为了避免这种情况并确保输入最重要的信息（而不是以后需要时可以检索的信息，例如主要来源甚至作者姓名），可以创建一个 SmartBlocks，首先鼓励对基本内容进行反思。为此，我们将使用命令 <%INPUT%> 打开一个对话框，在该对话框中，我们将被提示思考并直接输入基本思想。可以假设，这种提示比存在与相对标签相对应的属性等更有效。
Once we have identified the type of object and inserted the template, we need to complete it. The essential points in our examples of thought experiment, which will allow us to find them later, especially when we are not looking for it but it could be useful, are the relative tags (or concepts) and the relative questions or problems [12]. Completing these fields is also what requires the greatest intellectual effort. For convenience or lack of time, these fields are likely to be left blank or completed in a rush at the beginning. To avoid this and to ensure that the most important information is entered (rather than what can be retrieved later if needed, e.g., the primary source or even the author’s name), one can create a SmartBlocks that first encourages reflection on the essential. To do this, we will use the command <%INPUT%> to open a dialog box where we will be prompted to think and enter the essential ideas directly. It can be assumed that this prompt will be more effective than the presence, among others, of an attribute corresponding, for example, to relative tags.
为了便于创建与图形中现有元素（例如已在其他上下文中使用过的概念或问题）的链接，有两种策略：一种可以将输入字段设为自动完成字段，这样就可以找到图形中已经存在的页面。我们还可以自动运行查询，该查询将显示所有现有问题，我们可以在选择一些问题或创建新问题之前浏览这些问题。我们不会在这里介绍这个解决方案，但我们稍后会在“AoT 示例以保持与自己的对话”一节中讨论查询自动化。通过自动化该过程，可以鼓励人们考虑与其他元素的链接，从而利用图形结构。当然，没有自动化也是可能的，但您不会总是考虑它，特别是如果模板很复杂（如果有很多元素，那么要创建很多链接，您不会全部创建它们，而且通常会匆忙创建它们）。
To facilitate the creation of links with existing elements of the graph, such as concepts or questions already used in other contexts, there are two strategies: one can make the input field an autocomplete field, which allows to find pages already existing in the graph. We can also automatically run a query which will display all the existing questions, which we can browse before selecting some or creating new ones. We will not present this solution here, but we will talk about query automation later, in the section about “Examples of AoT to maintain a dialogue with oneself”. By automating the process, one is encouraged to consider the links with other elements and thus exploit the graph structure. It’s of course possible without automation, but you won’t always think about it, especially if the template is complex (if there are a lot of elements, so a lot of links to create, you won’t create them all, and often they will be created hastily).
为了实现这一点，一旦我们创建了新的知识对象，就会自动运行一个新的 SmartBlocks（“相关标签循环”）。它会提示搜索可能相关的现有标签（即页面）。然后，如果需要或没有匹配的标签，则创建新标签。AoT 具有递归循环的结构，它会无限期地调用自身，直到执行下一步。
To achieve this, a new SmartBlocks (“Related tag loop”) is automatically run once we have created a new knowledge object. It prompts to search existing tag (i.e., pages) that might be related. Then to create new tag if needed or if none of them matched. The AoT has the structure of a recursive loop, it calls itself indefinitely, until the next step is taken.
一个等效的、限制性稍小的提示（但因此对摆脱思维惯性的推动作用较小）可以通过提供一个按钮来实现，该按钮提示在每次添加新内容时重新开始搜索新参考资料。按钮可能比简单的静态提示更有激励作用，但比打开并要求搜索标签的对话框要少。当谈到养成习惯时，可能需要采用更基于激励的方法。安装后，一个简单的提示就足够了。
An equivalent, slightly less constraining prompt (but which therefore pushes a little less to get out of the inertia of one’s thinking) can be realized by providing a button that prompts to restart the search for a new reference at each new addition. A button is perhaps a little more incentive than a simple static prompt, but less than a dialog box that opens and asks to search for a tag. When it comes to forming a habit, the more incentive-based method may be necessary. When installed, a simple prompt is sufficient.
这些示例的主要目的是让人们理解 AoT 的逻辑，并初步了解 SmartBlocks 语法。自动化在这里提供的好处微乎其微：人们被引导到一个预定义的框架，并被鼓励首先思考记录信息的最重要元素。在这个阶段，我们已经开始掌握知识的本质、它与什么相关、阐明什么，以及它至少部分回答的问题。所有这些已经让它变得有意义，并让我们理解它。如此编织的关系网络对于将信息集合转化为知识至关重要，我们可以通过 PKG 功能轻松浏览。
The main purpose of these examples was to make understand the logic of AoTs and to give a first overview of the SmartBlocks syntax. The gain offered by automation here is minimal: one is directed to a predefined framework, and encouraged to think first about the most important elements to record the information. At this stage, we have started to grasp part of what is essential to a piece of knowledge, what it relates to and sheds light on, and the questions it at least partially answers. All this already makes it meaningful, and allows us to understand it. The network of relations so woven, that we can easily browse thanks to PKG features, is essential to the transformation of a collection of information into knowledge.
知识论证研究的 AoT 示例
Example of AoT for knowledge justification research
前面的例子可以涉及任何类型的知识，尤其适用于二手知识。对于我们确定的第二种知识类型，除了结构和与图中可用的其他信息的关系之外，最重要的是我们对它们合理性的认识，即我们首先为自己解释它们的能力，以及确定它们基础的局限性的能力。意识到这些局限性可以让我们重新思考并修改我们的一些知识。
The preceding examples can concern any type of knowledge and will be particularly appropriate for secondhand knowledge. The essential thing for the second type of knowledge we have identified is, more than the structure and the relations with other information available in the graph, the awareness we have of their justification, i.e., our capacity to give an account of them, first of all for ourselves, and to identify the limits of their grounding. Awareness of these limits is what can make us think again and revise some of our knowledge.
为了记录一个论点，我们可以简单地将“理由”属性添加到某种知识的模板中，并在需要时随时完成它。但同样，由于我们认为这是构成知识所必需的一种信息，并且经验表明，我们很容易满足于没有理由的信仰，而掌握一套我们认为可靠的理由往往是一项非常艰巨的任务，我们认为 AoT 可以很好地帮助激发所需的智力努力并在这方面锻炼我们自己。
To record an argument, we could simply add a “justifications” property to the template of a type of knowledge, and complete it whenever we want. But here again, since this is a type of information that we consider essential for the constitution of a knowledge, and since experience suggests that we are easily satisfied with believing without justification, and that grasping a set of justifications that we can consider solid is often a very demanding task, we think that an AoT can favorably help to trigger the required intellectual effort and to exercise ourselves in this regard.
首先，我们以开放式问题的形式向我们的对象类型添加一个属性来引发思考，例如“什么支持这个说法？”，并使用“添加参数”按钮来表明我们正在启动一个新的 AoT，即工作流程的一个子部分。
First, we add an attribute to our object type, in the form of an open-ended question to prompt thinking, e.g., “What supports this statement?”, and with an “Add an argument” button to indicate that we are initiating a new AoT, a sub-part of the workflow.
图 5.8 添加了两个参数，添加了第三个参数
Figure 5.8 Added two arguments, adding a third one
Smartblock 新论点将插入一个区块，我们将被邀请在其中制定论点，然后使用新按钮对其进行评估。“添加论点”按钮设置为多次点击，这将允许插入尽可能多的论点。
The Smartblock New argument will insert a block where we will be invited to formulate an argument, then to evaluate it with a new button. The “Add an argument” button is set up to be clicked several times, which will allow to insert as many arguments as necessary.
此外，Smartblock 新论点会显示一个“通知”，即一个弹出窗口消息，该消息会在一定时间（这里是 10 秒）后消失，其中显示一条建议，即一种方法的提醒，人们打算将其提供给未来的自己，而无需在每个新论点时用图表中的所有字母重写它。这里通知的文本直接写在 SmartBlock 中，但也可以链接到图表中的另一个位置，例如，人们收集了知识构建主要原则的页面。这些建议可能因对象的类型而异：例如，对于二手知识，人们首先在来源本身中寻找依据，如果找不到，人们再寻找其他来源，从而寻找其他二手知识。
Besides, the Smartblock New argument displays a “notification”, a message in a popup window that disappears after a certain time (here 10 seconds), which displays a recommendation, a reminder of a method, that one intends to give to one’s future self without rewriting it in all letters in the graph at each new argument. Here the text of the notification is written directly in the SmartBlock, but it could just as well be linked from another location in the graph, e.g., a page where one has gathered the main principles of the construction of one’s knowledge. These recommendations may vary according to the type of object: e.g., for secondhand knowledge, one first looks for the justifications in the source itself and, failing that, one looks for other sources, thus other secondhand knowledge.
SmartBlock 评估论证只是要求您选择您认为刚刚提出的论证所提供的论证程度。根据选择（在我们的示例中为证据、强、合理、弱），将插入相应的标签。这将使我们更容易找到有充分依据的断言（基于证据或有强论据支持的断言），以便在决策中依赖它们，或者找到更脆弱的断言（合理或弱），以便尝试巩固它们、使它们发展或简单地拒绝它们。最后，如果论证没有证据的强度（理论上不需要任何其他论证），我们建议启动新一轮的论证搜索循环：我们再次询问是什么支持了这一断言（我们刚刚提出的理由），我们将能够再次对其进行评估，依此类推，理想情况下直到我们得出一组证据。
The SmartBlock Evaluate an argument simply asks you to choose what you consider to be the degree of justification offered by the argument you have just formulated. Depending on the choice (evidence, strong, decent, weak in our example), a corresponding tag will be inserted. This will make it easier to find either the solidly justified assertions (evidence based or supported by strong arguments), in order to rely on them in our decisions, or the more fragile assertions (decent or weak), in order to try to consolidate them, to make them evolve or simply to reject them. Finally, if the argument does not have the strength of evidence (which does not require, theoretically, any other justification), we propose to initiate a new loop of search for justification: we ask again what supports this assertion (the justification we have just formulated), which we will be able to evaluate again, and so on, ideally until we arrive at a set of evidence.
当然，这个过程很少会进行到底。重要的是，我们开始这个过程，我们有机会反思现有的论证，并意识到我们所拥有的论证的质量。目标是训练我们的思维，使其倾向于寻找论证来巩固我们的知识。我们已经可以看到，这种算法将帮助我们建立一系列论证，并最终建立一个论证网络——PKG 的图形结构将使我们能够很容易地浏览这个网络——在这个网络中，我们将能够一目了然地看到（如果我们使用颜色或图标来直观地识别弱论证或强论证，就更是如此）哪些是我们可以放心依赖的，哪些是值得更远距离考虑并需要进一步检查的。从这个意义上说，建立扎实知识的 AoT 是一种在图形中定位自己的方式：由于这项工作，并非所有信息都是平等的，并非所有路径都无差别地通向知识。有些节点比其他节点是更好的通道点。
Of course, this process will rarely be carried out to its end. What is essential is that it is initiated, that we have had the opportunity to reflect on the justifications available, and that we are aware of the quality of the justifications we have. The goal is to discipline our mind to tend, in general, to look for justifications to consolidate our knowledge. We can already see that this algorithm will help us to build up series of arguments and, eventually, a network of arguments – that the graph structure of the PKGs will allow us to go through quite easily – within which we will be able to see at a glance (even more so if we play with colors or icons to visually identify the weak or strong arguments) which ones we can rely on with confidence and which ones deserve to be considered with more distance and require further examination. It is in this sense that AoTs to build up solid knowledge are a way to orient oneself in one’s graph: thanks to this work, not all information is equal, not all paths lead indifferently to knowledge. And some nodes are better points of passage than others.
评估一个人知识水平的 AoT 示例
Examples of AoT to evaluate one’s knowledge
评估刚刚获得或记录的知识当然并不容易。发现的乐趣或新理解的印象可能会导致对相关知识的可靠性的估计过高。为了弥补这种偏见，多次重新评估自己的知识（尤其是最重要的知识）是有用的。我们在下面提出了两个 AoT，邀请人们准确地重新评估自己的知识。
Evaluating knowledge that one has just acquired or recorded is certainly not easy. The pleasure of discovery or the impression of a new understanding can lead to overestimation of the soundness of the knowledge in question. To compensate for this bias, it is useful to reevaluate one’s knowledge, especially the most important ones, several times later. We propose below two AoTs that invite one to reevaluate one’s knowledge precisely.
用于比较信念程度和论证程度的 AoT 示例
Example of an AoT to compare degree of belief and degree of justification
它包括提出两个简单的问题，要求我们区分对信仰的执着程度和对其合理性的程度：
It consists in asking two simple questions that require us to distinguish between the degree of attachment to a belief and its degree of justification:
图 5.9 带日期条件和滑块的 SmartBlocks
Figure 5.9 SmartBlocks with condition on date and sliders
我对它有多大程度的相信？我们在这里问的是相信或信任的程度：我们在多大程度上相信这个命题是真的。这是一个观察一个人的依恋感的问题。这种依恋与发现的背景有关，它取决于我们的教育、我们的整个信仰和价值观，但也可能取决于简单的巧合、轶事关系。它主要由心理原因来解释，这些原因不一定是认为该断言为真的充分理由，但却促使我们相信它。在这方面，即使它没有说明命题的真实性，它也富含教训，特别是关于我们的价值观和我们的目的。
How much do I believe it? We are asking here about the degree of belief or credence: how much we believe this proposition to be true. It is a matter of observing one’s feeling of attachment. This attachment is linked to the context of discovery, it depends on our education, on the whole of our beliefs and values, but it can also depend on simple coincidences, on anecdotal relations. It is mainly explained by psychological causes, which are not necessarily sufficient reasons to hold the assertion as true, but which push us to believe it. In this, even if it says nothing about the truth of the proposition, it is rich in lessons, notably about our values and our ends.
它有多牢固？这里的问题是，有什么理由可用（或据称可用）。理由的程度取决于支持该断言为真的证据和理由，并且任何知道它的人都应该承认这些证据和理由。显然，这些证据和理由的质量，使人们或多或少关注它们、使人们或多或少重视它们的心理原因等，都涉及主观依恋的问题，而人们认为的事实可能是一个信念问题。
How solidly founded is it? The question here is what justification is available (or supposedly available). The degree of justification depends on the evidence and reasons for holding the assertion to be true, and which should be recognized by anyone else who is aware of it. Obviously, the quality of this evidence and reasons, the psychological reasons that make one pay more or less attention to them, that make one value them more or less, etc., refer to the problem of subjective attachment and what one holds as facts can be a matter of belief.
这两个维度之间的区分并不严格。但这里的练习是从两个不同的方向看，即一个人的感受和可用的数据。比较这两个视角可以很好地锻炼批判性思维[13]。批判性思维的目标之一是打造一种能够更好地识别客观依据的思维，并意识到有时会让我们滥用某些信念的心理或认知偏见。这是一个工作视野，一个人不可能完全客观（这就是为什么一个人需要科学方法和科学界来在需要客观性的领域更接近客观性，即使它仍然是一种理想）。
The separation between these two dimensions is not strict. But the exercise here is to look in two different directions, at one’s feelings and at the available data. Comparing these two perspectives can provide a good exercise in critical thinking [13]. One of the goals of critical thinking is to forge a mind that is better able to identify objective justifications and to become aware of the psychological or cognitive biases that sometimes attach us abusively to certain beliefs. It is an horizon of work, one cannot be perfectly objective by oneself (this is why one needs the scientific method and a scientific community to get closer to objectivity in the domains where it is desirable, even if it remains an ideal).
在这个 AoT 中，我们只是要求尽可能真诚地以 1 到 10 的等级评估依恋程度和合理性程度。两个量表之间的差距很大，如果信念高于合理性程度，我们会问自己是什么让我们相信它，涉及哪些价值观，等等。我们会被鼓励去寻找可能的合理性。相反，如果信念较低，我们会问自己为什么尽管有可用的数据，我们仍然不情愿。如果两个量表通常处于同一水平，那可能是因为我们无法区分这两个维度（或者不太可能，因为我们非常理性）。
In this AoT, we simply ask to evaluate, on a scale of 1 to 10, the degree of attachment and the degree of justification, as sincerely as possible. A gap between the two scales is significant, if the belief is higher than the degree of justification, we will ask ourselves what leads us to believe in it, which values are involved, etc. And we will be encouraged to look for possible justifications. If, conversely, the belief is lower, we will ask ourselves why we are reluctant despite the available data. If the two scales are most often at the same level, it is probably because we cannot distinguish between these two dimensions (or, less likely, because we are extremely rational).
显然，这两个维度只能事后评估，而不是在记录时进行评估。评估是根据经验进行的，特别是当这种断言受到质疑时我们的反应经验，或者我们试图解释或捍卫它的经验。观察这种评估的演变将很有启发性。为此，SmartBlock 会自动插入评估日期（如果您尚未进入每日笔记页面）。这样，就可以轻松地比较在不同时间点进行的评估并询问它们为什么发生了变化。然后可以重新评估可用的参数或添加新的参数。为了有机会重新评估我们的知识，我们可以依靠偶尔查阅相应的页面和视觉激励，例如红色标签，表示健全性为“#weak”。我们还将在“一些 AoT 示例以保持与自己的对话”部分中介绍其他方法来增加这些机会。
It is clear that these two dimensions can only be evaluated with hindsight and not at the moment of recording. The evaluation is made with experience, notably the experience of our reactions when such an assertion is called into question, or of our attempts to explain or defend it. It will then be instructive to observe the evolution of this evaluation. To do this, the SmartBlock automatically inserts the date of the evaluation (if you are not already on a daily note page). In this way, it will be easy to compare evaluations made at different moments in time and to ask why they have changed. The available arguments can then be reevaluated or new ones added. To have the opportunity to reevaluate our knowledge, we can rely on the occasional consultation of the corresponding pages and on visual incentives such as a red tag indicating that the soundness is “#weak”. We will also present other methods to multiply these opportunities in the section on some “Examples of AoT to maintain a dialogue with oneself”.
这里的评估本质上是一种元认知练习。最重要的是尝试去做，然后观察最终的演变。然后，你可以更新声明的“健全性”属性（例如，我们可以在四种状态之间切换：证据、强、体面和弱）。
The evaluation here is essentially a metacognition exercise. The most important thing is to try to do it, and then to observe the evolution eventually. Then, you can update the “Soundness” attribute of the claim (for example, we can switch between four states: evidence, strong, decent and weak).
用于综合知识的 AoT 示例
Example of an AoT to synthesize knowledge
通过努力综合所获得的知识，人们可以测试自己，以某种方式评估自己的知识，因为这种努力所代表的困难表明了人们对知识的理解或多或少肤浅。这不仅仅是以描述性的方式重复信息的问题，而是表达其逻辑的问题。尝试以简洁的方式表达它迫使人们更好地理解它。
By striving to produce a synthesis of acquired knowledge, one tests oneself, one evaluates one’s knowledge in a certain way because the difficulty that this effort represents indicates the more or less superficial understanding that one has of it. It is not simply a matter of repeating information in a descriptive way, but of expressing its logic. Trying to express it in a brief manner forces one to understand it better.
当然，这种综合可以在不遵循明确定义的步骤的情况下完成，也不一定需要 AoT。我们在此提出一个思想实验综合方法的例子，因为它是一种很容易被简化为二手知识的知识，因为它包含一些虚构的、轶事的、有时很有趣的故事，讲起来很愉快，但如果我们没有读过它的来源，我们就不一定能理解它。
This synthesis can of course be done without following well-defined steps and does not necessarily require AoT. We propose here an example of a synthesis method for thought experiments, because it is a type of knowledge that can easily be reduced to secondhand knowledge, since it consists in a little imaginary, anecdotal, sometimes funny story, that is pleasant to tell, but that we do not necessarily understand if we have not read the source from which it is coming.
感谢之前的模板，我们已经问过自己它们可以与哪些概念联系起来，它们可以帮助回答哪些问题，这已经很多了。但我们还没有充分说明它们如何用于推理，从而帮助我们思考。这就是我们打算用下面的 AoT 做的事情，它鼓励通过一系列问题更好地理解思想实验，并在这样做的过程中，写出它的综合的本质，我们可以在将来以比简单的信息描述更有益的方式参考它。这个 AoT 是对思想实验的个人反思的结果，它的逻辑当然值得怀疑，可能并不适用于所有的思想实验。
Thanks to the previous templates, we have asked ourselves to which concepts they could be linked, to which questions they could help answer, and that is already a lot. But we have not yet specified enough how they can be used in reasoning and thus help our thinking. This is what we propose to do with the following AoT, which encourages a better understanding of the thought experiment through a series of questions and, in doing so, writes the essence of its synthesis, to which we can refer in the future in a much more profitable way than to a simple informative description. This AoT is the result of a personal reflection on thought experiments, its logic is of course questionable and probably does not apply to all of them.
图 5.10 吉格斯环思想实验的综合：左侧的答案会自动插入右侧预先写好的结构中
Figure 5.10 Synthesis of the Gygès ring thought experiment: answers on the left are automatically inserted in the pre-written structure on the right
在询问这个思想实验是为了反驳还是支持某个主张之后（在模板的某些位置会插入“真”或“假”，如果是为了反驳某个主张，还会询问一个额外的问题：我们会问哪种理论更合适），我们会提出一系列问题，这些问题将鼓励我们更好地理解主导这个思想实验构思的逻辑，并以某种方式重构作者想要引导我们的推理（这通常是一个质疑的问题，将我们置于问题面前，这通常意味着表明我们经常认为显而易见的理论并不那么明显。）通过回答这些精确的问题，答案会自动插入（感谢 Roam 的块引用）到部分编写的综合结构中。显然，结果并不总是令人满意，可能需要重写，但这是一个练习，它给出了一个写作示例，最重要的是，它使人们掌握了激发这种写作的逻辑，同时有助于更好地理解思想实验。
After asking whether this thought experiment is designed to refute or support a claim (“true” or “false” will be inserted at certain points in the template, and an additional question will be asked if it is to refute a claim: we will ask which theory could be more appropriate), we will ask a series of questions that will encourage us to better understand the logic that presided over the conception of this thought experiment and to reconstitute in a way the reasoning to which the author wants to lead us (it is often a matter of questioning, of putting us in front of a problem, which often amounts to showing that a theory that we often hold to be obvious is not so much obvious.) By answering these precise questions, the answers are automatically inserted (thanks to Roam’s block references) into a partially written structure of the synthesis. Obviously the result will not always be satisfying, and it will probably be necessary to rewrite, but it is an exercise that gives an example of writing and above all that makes one grasp the logic that motivates this writing, while helping to better understand the thought experiment.
与自己保持对话的 AoT 示例
Examples of AoT to maintain a dialogue with oneself
拥有一套模板来构建和评估你的知识是一回事，定期使用它们是另一回事。为了给自己定期使用它们的机会，值得鼓励这样做。这可以通过链接或按钮来运行模板，以及通过不同的方式重新浮现你的知识来实现。当然，可以创建一个仪表板，其中列出了主要类型的知识，甚至可以使用数据库查询检索所有知识的列表，并列出可以应用于它们的所有 AoT，以便开发和评估它们。这通常是在经典的 PKM（个人知识管理器）中完成的，例如 Notion，它在 PKG 中也能很好地工作。这个仪表板页面是与我们的知识相关的所有节点的入口点。要利用它，我们必须打开相应的页面，浏览可能非常庞大的内容，并在提供的所有可能性中选择要做什么。
Having a set of templates to build and evaluate your knowledge is one thing, using them regularly is another. To give yourself the opportunity to regularly use them, it is worthwhile to be encouraged to do so. This can be done through the presence of links or buttons to run a template, and through the different ways to resurface your knowledge. Of course, it is possible to create a dashboard where the main types of knowledge are listed, or even to retrieve a list of all the knowledge with a database query, and to list all the AoTs that can be applied to them in order to develop and evaluate them. This is what is usually done in classic PKMs (personal knowledge manager) like Notion and it also works well in a PKG. This dashboard page is then the entry point to all the nodes related to our knowledge. To take advantage of it we have to open the corresponding page, browse the contents, which may be vast, and choose what to do among all the possibilities offered.
下面我们介绍另一种为自己的知识提供切入点的方法，这种方法由 Roam Research 推广，现在大多数 PKG 中都有：Roam Research UX 的主要选择之一是，打开应用程序时，我们会进入一个空白页，其标题是当前日期：每日笔记页面 (DNP)。DNP 系列是所有笔记的支柱，我们可以从中创建指向我们添加到图表中的内容的链接，以利用时间上下文来排序和找到我们的想法。
Below we describe another way to give yourself an entry point to your knowledge, which has been popularized by Roam Research and is now present in most PKGs: one of the major choices of Roam Research’s UX is that when opening the application, we land on a blank page whose title is the current date: the daily note page (DNP). The DNP series is the backbone of all the notes, from which we can create links to the content we add to our graph to take advantage of the temporal context to order and find our ideas.
每日模板
Daily template
但是，您不必每天从空白的 DNP 开始，而是可以插入一个模板（可能由于 SmartBlocks 扩展中的一个选项而自动插入），该模板为当天提供了框架。我们将其称为每日模板，它在那些在 PKG 中每天写日记的人中非常受欢迎。我们将看到，借助 SmartBlocks 等可编程模板，可以根据星期几等情况显示不同的模板，并集成按钮，使我们能够轻松执行其他模板或重新显示一组要处理的图表节点。因此，每一天都会提供一组访问图表不同节点的访问点，以及尽可能多的浏览它或找到它的某些角落的可能性。该系统提供了一种与自己保持对话的方式，讨论您过去写下的想法，或考虑您为未来的自己提出的建议或要求。
However, instead of starting each day with a blank DNP, you can insert a template (possibly automatically thanks to an option in the SmartBlocks extension) that provides a framework for the day. We will refer to this as the daily template, which is very popular among those who practice daily journaling in PKGs. We will see that thanks to the programmable templates like SmartBlocks, it is possible to make a different template appear, depending, for example, on the day of the week, and to integrate buttons that allow us to easily execute other templates or to resurface a set of nodes of the graph to work on. Thus, each day will offer a set of access points to different nodes of the graph, and as many possibilities to browse it or to find certain corners of it. This system offers a way to maintain a dialogue with yourself, discussing thoughts you have written down in the past, or considering advice or requests you have made for your future self.
为了记录新知识，我们可以插入一个问题，例如“我今天学到了什么？”，并添加一个固定按钮，该按钮将提出我们已经看到的现有知识类型的列表或添加新类型，并且将为每个新的知识位创建一个新块，引用为其创建的新页面。
To record new knowledge, we can insert a question like “What did I learn today?” and add a fixed button that will propose, as we have already seen, a list of the existing types of knowledge or the addition of a new type, and that will create a new block for each new bit knowledge, referring to the new page created for it.
但最有趣的是可以提出图表的节点来重新发现它们并有利于偶然发现，或者评估和丰富它们，或者与自己展开批判性对话。
But most interesting is the possibility of bringing up nodes of the graph to rediscover them and favor serendipity, or to evaluate and enrich them, or to initiate a critical dialogue with yourself.
这里的目标不是涵盖可以集成到日常模板中的所有内容，而是重点介绍集成我们在前面章节中介绍的模板的示例。
The goal here is not to cover everything that can be integrated into a daily template but to focus on examples to integrate the templates that we have presented in the previous sections.
将目标数据集带到表面
Bringing a targeted data set to the surface
有几种方法可以显示以前记录的知识，例如通过嵌入专门为此目的创建的页面或块，或者通过查询数据库。
There are several ways of surfacing previously recorded knowledge, such as by embedding pages or blocks that have been specifically created for this purpose, or by querying the database.
当然，最简单的方法就是插入对目标知识类型的引用，这样用户就可以快速浏览它们。在大多数管理反向链接的 PKG 系统中，您可以轻松插入引用给定节点的所有节点的列表，例如引用“声明”或“健康”的所有页面。然后可以使用关键字过滤这些数据以包含或排除，这使您能够利用您建立的不同类型的知识及其属性相关的值，特别是“相关标签”和“相关问题或问题”。例如，在 Roam Research 中，如果我们插入对“声明”页面的所有引用（带有 {{[[mentions]]:[[Claim]]}}），我们可以对其进行过滤以仅显示与“弱”标签相关的声明。
The easiest way, of course, would be to simply insert references to targeted types of knowledge, which allow users to browse them quickly. In most PKG systems that manage backlinks, you can easily insert a list of all nodes that refer to a given node, for example all pages that refer to “Claim” or “Health”. It is then possible to filter this data with keywords to include or exclude, which allows you to take advantage of the different types of knowledge you built up and the values associated with their attributes, in particular “related tags” and “related questions or problems”. For example, in Roam Research, if we insert all the references to the “Claim” page (with {{[[mentions]]:[[Claim]]}}), we can filter it to show only the claims related to “weak” tag.
图 5.11 对“主张”的引用，仅提及包含标签“弱”的主张
Figure 5.11 References to “Claim”, only mentions that include the tag “weak”
如果你拥有大量的知识，而只想在有限的知识范围内进行工作，那么你可以创建一个页面，其中包含你在特定时间最关心的五个问题的所有链接参考，并将此页面嵌入到每日模板中，以快速浏览相关知识。
If you have a large body of knowledge and you want to work on a limited set, you could, for example, create a page with all the linked references to the five questions that concern you the most at a given time, and embed this page in the daily template, to quickly go through the related knowledge.
提取目标数据部分的另一种方法是根据多个逻辑相关的标准执行查询（例如，所有我们找到“索赔”和“健康”的块 - 至少通过层次结构间接找到）。查询显然是一种经典的数据库功能。在 PKG 中，正如我们在第 XX 章关于非故意关系中所述，兴趣在于这些关系更灵活且易于建立，然后我们可以轻松浏览它们，或者插入元素而不受固定框架的限制。
Another way to extract a targeted part of the data is to perform queries according to multiple logically related criteria (e.g., all blocks in which we find “Claim” AND “Health” – at least indirectly, by the hierarchical structure). Queries are obviously a classical database functionality. In a PKG, the interest, as we developed in the chapter XX on unintentional relations, is that these relations are more flexible and easy to establish, and that we can then easily browse them, or insert elements without the constraints of the fixed framework.
我们可以简化和自动化这个过程（因为查询是用必须学习的语法编写的 [14]），方法是使用 SmartBlock 根据提供的条件进行查询。在我们的示例“查询 A 和 B”中，我们提出了一个简单的查询，即两个值（A 和 B）的结合，变量可以手动设置（通过输入框）或在模板中自动设置。例如，这允许使用同一个 SmartBlock 在一天对 A 和 B 运行查询，在另一天对 C 和 D 运行查询。我们将在下一节中看到如何使 SmartBlock 依赖于星期几。
We can simplify and automate this process (because the queries are written in a syntax that must be learned [14]) by using a SmartBlock that will make the query according to the criteria provided. In our example “Query on A AND B”, we propose a simple query, the conjunction of two values (A AND B), with variables that can be set manually (via an input box) or automatically in the template. That allows, for example, using the same SmartBlock, to on one day run a query on A and B, and on another day on C and D. We will see in the next section how to make a SmartBlock dependent on the day of the week.
将随机数据集带到表面
Bringing a random data set to the surface
随机检索知识也很有用。每天或每周留出一些时间来复习一些知识总是一个好主意，但这样做的风险是，你可能会因信息量太大而灰心丧气，你可能会在不知不觉中专注于一些知识而忽略其他知识。定期随机地提出有限数量的元素可以避免这些问题。[15]
Randomly retrieving knowledge can also be very useful. It is always a good idea to set aside some time during the day or week to review some of your knowledge, but there is a risk that you will be discouraged by the amount of information, and that you will focus on some and neglect others without being aware of it. Bringing up a limited number of elements randomly on a regular basis avoids these issues. [15]
必须考虑那些我们不会自发寻找的想法，或者我们倾向于拒绝的想法，这可能是决定性的。达尔文已经为这种做法提供了这样的地位：
Having to consider ideas that we would not have spontaneously looked for, or that we would have tended to reject, can be decisive. Here is the place that Darwin already gives to such a practice:
多年来，我还一直遵循着一条金科玉律，那就是，每当我遇到一个已发表的事实、新的观察或想法，与我的总体结论相反时，一定要立即记下来；因为我的经验告诉我，这些事实和想法比有利的事实和想法更容易被遗忘。由于这个习惯，很少有人对我的观点提出异议，我至少会注意到并试图回答。[16]
I had, also, during many years followed a golden rule, namely, that whenever a published fact, a new observation or thought came across me, which was opposed to my general results, to make a memorandum of it without fail and at once; for I had found by experience that such facts and thoughts were far more apt to escape from the memory than favourable ones. Owing to this habit, very few objections were raised against my views which I had not at least noticed and attempted to answer. [16]
与传统的孤岛结构相比，PKG 的结构已经大大增加了考虑未曾考虑过的想法的机会。AoT 提供的自动化功能可以进一步增强这种机会。一旦某个元素随机出现，例如某个声明，我们就可以轻松应用前面看到的评估 AoT：“评估陈述的可靠性”。作为智能块，它可以自动插入为随机声明的子块，因此您可以直接对其进行评估。
The opportunity to consider ideas that one was not looking for is already greatly increased by the structure of PKGs as opposed to classical silo structures. It can be further enhanced by the automation offered by AoTs. Once an element is randomly resurfaced, for example a claim, we can easily apply the evaluation AoT seen earlier: “Evaluate reliability of a statement”. As a Smartblock, it can be automatically inserted as a child block of the random claim, so you can directly evaluate it.
图 5.12 SmartBlocks 嵌入随机声明并运行评估 SmartBlocks
Figure 5.12 SmartBlocks embedding a random claim and running the evaluation SmartBlocks
我们必须承认，此功能尚未完全开发，有时会显示不相关的块，具体取决于我们引用它们的方式。此外，它没有提供足够精确地定位我们想要随机显示的块类型的方法。但这肯定是一个值得未来开发的功能，并且无疑最有可能激发与自己的对话，尤其是如果它除了这里提出的评估之外，还与其他类型的互动相结合，例如受波普尔启发的批判性问题（例如，“应该向我展示什么证据才能让我放弃这种信念？”）或批判性思维练习。
We have to admit that this functionality is not yet fully developed and sometimes brings up irrelevant blocks, depending on the way we reference them. Besides, it does not offer the means to target precisely enough the type of block we would like to bring up randomly. But it is certainly a feature that would be interesting to develop in the future and which is undoubtedly the most likely to stimulate a dialogue with oneself, all the more so if it is combined, in addition to the evaluation proposed here, with other types of interaction, such as Popperian-inspired critical questions (for example, “What evidence should be presented to me for me to give up this belief?”) or critical thinking exercises.
自动化和人工智能与生命和创造性思维相比？
Automation and AI versus living and creative thinking?
AoT 不是一个固定的框架。理想情况下，它们是我们自己反思知识过程的产物，我们可以通过经验来发展它们。而且，没有什么强迫我们遵循它们。这里自动化的不是思想本身，而是以我们认为在特定时刻好的方式对一系列思想进行排序的激励。我们仍然需要积极思考。而不是拖延或从头开始重新思考，没有方法，这是一个更好的机会，让我们努力以一种被认为有效和必要的方式思考。
AoTs are not a fixed framework. Ideally, they are the product of our own reflection on our knowledge processes, and we can evolve them with experience. Moreover, nothing forces us to follow them. What is automated here is not the thought itself, but the incitement to order a sequence of thoughts in a way that we consider good at a given moment. We are still required to think actively. And rather than procrastinating or rethinking from scratch, without method, here is a better opportunity to make the effort to think in a way found to be effective and seen as essential.
尽管如此，人工智能可以充分利用图结构的复杂性。重要的联系可以隐藏起来。例如，也许我们的图中有一个论点可以支持和完成我们当前正在捕捉的论点，或者相反，也许有一个论点反对它并需要进一步质疑。我们可以想象一个可以识别和建议此类联系的人工智能。如果它停留在建议的水平，只要它们是可选的，人工智能就不会取代活生生的思想，而是可以刺激它。
Nevertheless, the complexity of the graph structure could be favorably exploited by an AI. Important connections can remain hidden. For example, maybe there is an argument in our graph that could support and complete the argument we are currently capturing, or on the contrary, perhaps one that opposes it and requires additional questioning. One can imagine an AI that could identify and suggest such links. If its remains at the level of suggestions, and as long as they are optional, an AI does not replace living thought and can stimulate it.
人工智能还可以提出相关问题，类似于我们之前在 AoT 中设置的问题，但它可能会改变这些问题，以打破常规，打破我们已经知道模板中会出现哪些问题的习惯。人工智能可以要求额外的理由，提出问题或提出异议。这在今天很容易实现，例如，通过使用 OpenAI 的 GPT-3。同样，在人类仍然“参与其中”并处于关键阶段的情况下进行思考可能会有益。事实上，即使假设思考行为并不重要，但结果可能更重要，我们相信，如果知识不是主动思考的产物，那么它就不是真正整合的，如果知识不能被创造性思维超越，那么它就只有一半的用处。人工智能替我们思考的风险在于，虽然它让我们只需点击几下鼠标就能获得越来越多的信息，但它也剥夺了我们真正的知识，这些知识我们不仅可以保存在记忆中，还可以激活，并借助这些知识做出判断、决策和创新。
An AI could also ask relevant questions, similar to the ones we set up in the AoTs presented previously, but it might vary them to break the routine and the tendency to skip when we already know which question is going to be in the template. An AI can ask for an additional justification, ask a question, or raise an objection. This is something that is easily achievable today, for example, by using GPT-3 from OpenAI. Again, it could be beneficial to thinking with a human still “in the loop” and at the essential stage. Indeed, even assuming that the act of thinking would not be essential but that the result may count more, we believe that a piece of knowledge is not really integrated if it is not the product of an active thinking, and that knowledge is only half useful if it cannot be transcended by a creative thinking. The risk of an AI that thinks for us is, while making more and more information available at a fewer clicks, that it deprives us of real knowledge that we cannot only possess in a memory, but that we can activate and thanks to which we can make judgments, decisions and innovations.
结论
Conclusion
PKG 激发了这种思考，我们相信，通过根据我们的需求和目的量身定制的 AoT，少量深思熟虑的自动化可以进一步激发积极的思考，而不会脱离它所坚持的真理。根据我们想要到达的目的地，即我们想要找到或检查的知识，我们可以轻松获得向我们展示图表中可能的相关路径的工具。例如，我们将能够快速导航到我们在某个领域最成熟的知识，并快速概览可用的依据，以做出有根据的决策。或者，相反，我们可以导航到最脆弱的信念，重新评估它们的理由，最终放弃它们并认真对待其他立场。
PKGs stimulate such thinking, and we believe that a small amount of thoughtful automation, through AoTs that we tailor to our needs and purpose, can further stimulate active thinking that doesn’t become disconnected from what it holds as true. Depending on the destination we want to reach, i.e., the knowledge we want to find or examine, we can easily acquire tools that show us possible relevant paths within our graph. For example, we will be able to quickly navigate to our best-established knowledge in a domain and get a quick overview of the available justifications to make well-founded decisions. Or, on the contrary, we can navigate towards the most fragile beliefs, reevaluate their justifications and eventually abandon them and take alternative positions seriously.
通过为我们的图添加一些结构和自动化，我们不会失去它的灵活性。而且，我们更好地利用了它的主要优势之一，即内容的共存，否则这些内容将保持惰性，并通过它们的相遇产生更活跃和更敏锐的知识。随着 PKG 的增长，它变得越来越复杂。增加成千上万个节点之间的链接有时会给人一种迷失方向甚至迷失的印象。好的搜索引擎可以部分缓解这个问题。但我们只寻找我们知道存在的东西，并且遵循我们当前思维趋势的东西。好的 AoT 可以帮助我们在复杂的 PKG 中找到最佳路径，这些路径不一定经过我们最熟悉的节点，也不一定是最短的路径，但它们的设计是为了最大限度地发挥我们的智慧。
By adding a little structure and automation to our graph we do not lose its flexibility. And we take better advantage of one of its main strengths, the concomitance of contents that would otherwise have remained inert, and that produce, by their meeting, a more active and sharpened knowledge. As it grows, a PKG becomes more complex. Multiplying the links between tens of thousands of nodes can sometimes give the impression of being disoriented or even lost. Good search engines can partially alleviate this problem. But we only look for what we know exists and which follows the current trend of our thinking. Good AoTs can help us find the best paths through the complexity of our PKG, paths that will not necessarily go through the nodes we are most familiar with, or that will not necessarily be the shortest, but that are designed to make the most of our intelligence.
笔记
Notes
[1] PKG 将用于指代图表以及创建和管理图表的工具。当谈到“他或她的”PKG 时，我们会假设它是一个单一实体，但一个人当然可以管理用作 PKG 的多个图表（为了提高研究的精确度，但我们会在横向性方面有所损失）。
[1] PKG will be used to refer to both the graph and the tool to create and manage it. When talking about “his or her” PKG, we will suppose that it is a single entity, but a person can of course manage several graphs which are used as a PKG (for a gain of precision in the research, but we then lose in transversality).
[2] 当然，这只是众多知识概念中一种可能且有争议的概念。我们感兴趣的是证明我们信念的必要性，因为我们想表明，我们可以利用 PKG 制定某些策略来加强我们信念的证明。有关柏拉图的定义及其提出的问题的完整讨论，请参阅：Ichikawa, J. 和 Steup, M. (2017)《知识分析》。《斯坦福哲学百科全书》。https://plato.stanford.edu/entries/knowledge-analysis
[2] This is of course only one possible and controversial conception of knowledge among others. It is the need to justify our beliefs that interests us here, insofar as we want to show that we can put in place certain strategies to strengthen the justifications of our beliefs by taking advantage of PKGs. For a complete discussion of Plato’s definition and the problems it raises, see: Ichikawa, J. and Steup, M. (2017) The Analysis of Knowledge. In Stanford Encyclopedia of Philosophy. https://plato.stanford.edu/entries/knowledge-analysis
[3] 当然，如果作者分享了论点或证据，更重要的是，如果他们尊重科学方法，他们的研究结果已经过同行评审，并且达成了一定的共识，那么将它们以我们在下一节中理解的强意义整合到我们的知识中将是合理的。
[3] Of course, if the author shares arguments or evidence and, even more so, if they respect a scientific method, their results have been reviewed by peers, and there is a certain consensus, it will be reasonable to integrate them into our knowledge in the strong sense that we understand it in the following section.
[4] Reichenbach, H. (1938) 经验与预测。知识的基础和结构分析。芝加哥：芝加哥大学出版社。
[4] Reichenbach, H. (1938) Experience and Prediction. An Analysis of the Foundations and the Structure of Knowledge. Chicago: University of Chicago Press.
[5] 请参阅 Cortex Futura (2021)，真正的思维算法已经到来。https://www.cortexfutura.com/algorithms-of-thought-have-arrived/
[5] See Cortex Futura (2021), True Algorithms of Thought Have Arrived. https://www.cortexfutura.com/algorithms-of-thought-have-arrived/
[6] 威廉·詹姆斯（1890），《心理学原理》，第 11 章
[6] William James (1890), The principles of psychology, Chap. XI
[7] 最近的两个 PKG Tana 和 Capacities 仍在开发中，它们在图结构中提供结构化数据的本机管理，类似于传统关系数据库中的管理。
[7] Two recent PKGs Tana and Capacities, which are still under development, offer native management of structured data, similar to that found in classical relational databases, within a graph structure.
[8] 因为这是我们最熟悉的 PKG 系统，而且我们将利用专为 Roam Research 设计的 SmartBlocks 扩展来构建可编程模板。并非所有考虑的功能都可以直接移植到其他 PKG（如 Obsidian、Logseq 或 Tana），但主要部分应该是可适应的。
[8] Because it’s the PKG system we have the most experience with, and because we’re going to take advantage of the SmartBlocks extension, designed for Roam Research to build programmable templates. Not all the features considered will be directly transposable to other PKGs like Obsidian, Logseq or Tana, but the main part should be adaptable.
[9] SmartBlocks 是可编程模板，目前有近 80 个命令用于预填充或根据上下文调整模板。该扩展最初由 TfTHacker 开发，后来由 David Vargas 进行了大幅升级。如果您想测试或修改此处介绍的 SmartBlocks，可以从 RoamJS SmartBlocks 扩展的 SmartBlocks Store（https://roamjs.com/extensions/smartblocks/）安装它们，名称为“PKG Book SmartBlocks 示例”。
[9] SmartBlocks are programmable templates with currently nearly 80 commands to prepopulate or adapt the template to the context. The extension was initially developed by TfTHacker then considerably upgraded by David Vargas. If you want to test or modify the SmartBlocks presented here, they can be installed from the SmartBlocks Store of the RoamJS SmartBlocks extension (https://roamjs.com/extensions/smartblocks/), under the name “PKG Book SmartBlocks examples”.
[10] Roam Research 有一个很棒的扩展，即由 RoamJS 开发的 Discourse Graph（https://roamjs.com/extensions/discourse-graph），它允许我们管理声明、证据和问题的表达，并有助于产生综合。我们稍后将介绍的大部分内容都可以用它完成，甚至更多。我们决定不使用它，以免给 SmartBlocks 的演示增加技术复杂性，因为 SmartBlocks 已经相当技术化了。但是，尽管它主要是由马里兰大学信息研究学院的 Joel Chan 为学术界设计的，但它对于个人知识来说非常有用且适应性强。
[10] There is a remarkable extension for Roam Research, Discourse Graph by RoamJS (https://roamjs.com/extensions/discourse-graph), which allows us to manage the articulation of claims, evidence and questions, and helps to produce synthesis. A large part of what we will present later could be accomplished with it and much more. We decided not to use it in order not to add technical complications to the presentation of SmartBlocks, which is already quite technical. But, although it is primarily designed for the academic community by Joel Chan, from the University of Maryland’s College of Information Studies, it is quite usable and adaptable for personal knowledge.
[11] 标签不过是带有 # 符号的引用，是提及页面、图表节点的另一种方式。
[11] A tag is nothing else than a reference preceded by a #, which is just another way of mentioning a page, a node of the graph.
[12] 还有其他解决方案可以帮助重现。“AoT 与自我保持对话的示例”部分将介绍其中几种。
[12] There are other solutions to help with resurfacing. Several will be presented in the section “Examples of AoT to maintain a dialogue with oneself”.
[13] 批判性思维是一套形成更理性、更有根据的判断的方法，并培养相应的技能，如逻辑准确性、对不同观点的开放态度和质疑自己观点的能力，或评估和识别证据的能力。一般来说，AoT 可用于培养批判性思维。无论如何，这正是我们在本章中提出的所有方法的目的。我们将它们视为评估我们的想法并引导我们走向最合适的想法的一种方式。
[13] Critical thinking is a set of methods to form more rational and well-founded judgments, and to develop the corresponding skills, such as logical accuracy, openness to different points of view and to questioning one’s own opinions, or the ability to evaluate and recognize evidence. AoTs in general can be used to cultivate critical thinking. This is in anyway the purpose of all the ones we propose in this chapter. We see them as a way to evaluate our thoughts and to direct us towards the most appropriate ones.
[14] 还有一些扩展可以在不了解语法的情况下生成可能非常复杂的查询，例如 Roam Research 的 Query Builder。Tana 本身就提供了这样的界面，这极大地方便了查询的创建（这里称为“实时搜索”）。
[14] There are also extensions to generate potentially very complex queries without knowing the syntax, like Query Builder for Roam Research. Tana offers such an interface natively, which greatly facilitates the creation of queries (called “Live Search” here).
[15] 每日复习可能是养成习惯的好方法。借助 SmartBlocks 的条件命令，可以想象每周的每一天都会重新出现不同类型的知识或领域。只需测试一周中的哪一天（使用“IFVAR”和“DATE”命令），并根据日期为一个或多个变量分配不同的内容，并可能根据可用时间重新出现可变数量的项目。
[15] A daily review is probably a good practice to build the habit. With the conditional commands of SmartBlocks, one can imagine having different types of knowledge or domains resurfaced each day of the week. It would be just a matter of testing the day of the week (with “IFVAR” and “DATE” commands) and assigning one or more variables different contents depending on the day, and possibly a variable number of items resurfaced depending on the time available.
[16] Darwin, C.《查尔斯·达尔文的生平和书信》，第一卷，第 36 页。https://charles-darwin.classic-literature.co.uk/the-life-and-letters-of-charles-darwin-volume-i/ebook-page-36.asp
[16] Darwin, C. The Life and Letters of Charles Darwin, Vol I, p.36. https://charles-darwin.classic-literature.co.uk/the-life-and-letters-of-charles-darwin-volume-i/ebook-page-36.asp
第六章
Chapter 6
利用 SEN 将桌面扩展为个人知识图谱
Extending the Desktop into a Personal Knowledge Graph with SEN
格雷戈尔·罗森瑙尔
近年来，支持链接和标记的个人笔记工具不断涌现，无论是基于 Web 的解决方案（如 Notion 或 Roam），还是桌面解决方案（如 Obsidian 或 Logseq）。所有这些工具都人为地将我们在日常工作流程中收集的信息与我们想要保留、丰富和搜索的信息区分开来。它们还会导致媒体中断，迫使用户进入自己的生态系统，从而导致重复并减少收集信息的重复使用。
Recent years have seen the rise of personal note-taking tools with support for linking and tagging, either web-based solutions like Notion or Roam, or desktop solutions like Obsidian or Logseq. All these tools introduce an artificial separation between information we collect as part of our everyday workflow and information we want to keep, enrich, and search. They also cause a media break and force users into their own ecosystem, leading to duplication and reducing reuse of collected information.
这违背了个人做笔记的目的：收集见解和知识，有助于发展想法并将收集到的知识应用于实际项目。
This counteracts the goal of personal note-taking: gathering insights and knowledge, which helps develop ideas and apply the knowledge gathered in actual projects.
个人知识图谱不应被外部化或由专门的应用程序控制，而应成为用户通用桌面环境的一个组成部分，旨在直接操作用户数据。
A personal knowledge graph should not be externalised or controlled by specialised applications but be an integral part of the user’s common desktop environment, which is designed for direct manipulation of user data.
这种解决方案过去曾被尝试过，称为“语义桌面”，但未能获得用户采用。早期的尝试完全采用了 OWL 和 RDF 等语义网标准，导致大量资源开销，得不偿失。
This solution has been tried in the past, known as “semantic desktop”, but failed to gain user adoption. Early attempts fully embraced semantic web standards like OWL and RDF, causing a significant resource overhead that was not worth the benefits.
本章介绍了 SEN（语义扩展），它是开源桌面操作系统 Haiku 的轻量级但功能强大的语义桌面层，使用现代文件系统来存储用户信息的关系和属性。
This chapter introduces SEN (Semantic ExteNsions), a lightweight but powerful semantic desktop layer for the open source desktop OS Haiku, using a modern filesystem to store relations and properties of user information.
只需对文件浏览器进行少量扩展，用户就可以直观地浏览文件及其关系。SEN 采用以数据为中心的方法，其中文件代表任何类型的实体，包括电子邮件、联系人或现实世界中的对象，如书籍、食谱或位置。使用 SEN，应用程序可以集成到知识图中。作为示例，将显示一个通用笔记本 (UNO)，它提供了一个简单的语义文本编辑器，编辑器内部和外部（使用文件浏览器）都有实体链接和导航，以形成数字第二大脑或“Zettelkasten”。
With minimal extensions to the file browser, users can intuitively navigate files and their relations. SEN embraces a data-centric approach, where files represent any kind of entity, including emails, contacts, or real-world objects like books, recipes or locations. Using SEN, applications can integrate into the knowledge graph. As an example, a universal notebook (UNO) will be shown, providing a simple semantic text editor with entity linking and navigation, inside and outside the editor (using the file browser), to form a digital second brain or “Zettelkasten”.
动机和愿景声明
Motivation and Vision Statement
本章介绍了 SEN（“语义扩展”的缩写），这是一个私人（即将开源）研究项目，它有一个真实世界的原型，用于实现最字面意义上的个人知识图谱系统：数字“第二大脑” (参见 Forte，2022)，嵌入到个人桌面操作系统中。
This chapter introduces SEN (short for “Semantic ExteNsions”), a private (and soon to be open sourced) research project with a real-world prototype for realising a personal knowledge graph system in its most literal sense: a digital “second brain” (see also Forte, 2022), that is embedded into a personal desktop operating system.
通过精心地提供对众所周知的桌面隐喻和文件系统的扩展，SEN 将任何文件（无论是文档、音频、视频还是简单的文本注释）连接到几乎任何其他类型的相关数据，不仅可以存储和分类信息（想想标签），还可以将事物置于透视图中并使关系可见且可导航。
By carefully providing extensions to well-known desktop metaphors and to the file system, SEN connects any file (be it a document, audio, video or a simple text note) to practically any other kind of related data, not only storing and categorising information (think of labels), but also putting things into perspective and making relations visible and navigable.
SEN 无需引入另一个专门用于笔记记录和链接的应用程序即可实现所有这些功能，而是仅对磁盘上存储的数据进行操作。这样一来，用户就可以直观、透明地处理和连接个人数据。在此过程中，他们会自然而然地从一组由元数据链接的文件中创建个人知识图谱，这些文件可以自动提取或手动添加。
SEN does all that without introducing yet another specialised application for note-taking and linking, but instead operates solely on the data stored on disk. This allows users to work with and connect their personal data intuitively and transparently. In the process, they naturally create a personal knowledge graph from a collection of files linked by metadata, which may be extracted automatically or added manually.
这导致了与众多知识管理系统的另一个重要区别：由于 SEN 集成到桌面并在文件系统级别工作，因此所有数据都会自动成为个人知识图谱的潜在节点，而无需用户在单独的工具中重新获取信息。通过采用通用的“一切都是文件”方法，SEN 提供了一种简单但功能强大、直观且不引人注目的解决方案，用于将链接信息作为日常工作流程的一部分进行管理。
This leads to another important differentiation from the myriad of knowledge management systems: because SEN is integrated into the desktop and works at the file system level, all data automatically becomes a potential node of a personal knowledge graph, without requiring users to recapture information in a separate tool. By embracing a universal “everything is a file” approach, SEN provides a simple, but powerful, intuitive and unobtrusive solution for managing linked information as part of everyday workflow.
因此，SEN 主要设计用于个人和离线使用，但由于它操作普通文件，因此个人知识图谱可以像普通文件或文件夹一样共享。[1]
Consequently, SEN is primarily designed for personal and offline use, but since it operates on ordinary files, a personal knowledge graph can be shared just like normal files or folders. [1]
背景和现状的批判性审视
Some Background and a Critical Examination of the Current State
个人知识图谱背后的想法可以追溯到 20 世纪 60 年代，当时 Luhmann 提出了如今著名的 Zettelkasten 方法（Schmidt，2018 年），甚至可以追溯到 1945 年，当时 Vannevar Bush 在他的开创性论文《正如我们所想》（Bush，1945 年）中介绍了一种名为“Memex”的个人知识管理系统。然而，直到最近几年，随着 Notion 或 Roam（商业）等联网数字笔记工具的推出，或 Foam（foambubble，2022 年）、Logseq 或 Obsidian 等开源替代品的推出，这个想法才变得越来越流行。
The idea behind personal knowledge graphs dates to the 1960s with the now famous Zettelkasten method by Luhmann (Schmidt, 2018), and even back to 1945, when Vannevar Bush introduced a personal knowledge management system called “Memex” in his seminal essay “As we may think” (Bush, 1945). However, only in recent years has this idea become increasingly popular following the introduction of connected digital note-taking tools like Notion or Roam (commercial), or open source alternatives like Foam (foambubble, 2022), Logseq or Obsidian, just to name a few.
然而，个人知识图谱解决方案的实用性也受到了合理的批评（Murphy，2022），因为简单地将注释连接到图中并不一定能提供有价值的见解。当前此类系统主要使用标记，这是最简单的分类形式。这不会传达任何更深层次的语义以允许对实体进行有意义的识别。即使提供简单的链接也不足以表达某些信息是如何以及为什么连接的。如果用户无法浏览和查询收集到的所有信息以及形成的任何连接，就无法收集有意义的知识。
However, there is also valid criticism of the usefulness of personal knowledge graph solutions (Murphy, 2022), as simply connecting notes in a graph does not necessarily provide valuable insights. Current systems of this kind mainly use tagging, which is the simplest form of classification. This does not convey any deeper semantics to allow meaningful identification of entities. Even providing simple linking is not sufficient to express how and why some bits of information are connected. Users cannot gather meaningful knowledge without being able to navigate and query all the information gathered and any connections formed.
许多笔记应用程序还限制了可以捕获的信息类型，并对学习和维护个人知识图谱提出了额外的要求。例如，基于 Web 的解决方案虽然已经变得非常复杂，但限制了用户处理和连接数据的方式。由于与单一供应商绑定，数据可能随时无法访问或使用。
Many note-taking applications also limit the kind of information that can be captured and pose additional requirements in learning and maintaining a personal knowledge graph. For example, web-based solutions, as sophisticated as they may have become, limit how users can work with and connect their data. By being bound to a single vendor, the data may become inaccessible or unusable at any time.
云提供商要么收取经常性费用，要么利用用户数据牟利。这甚至可能对不安全环境中的用户（如记者或活动家）构成安全威胁，并且不适合敏感数据或商业数据（可能侵犯知识产权、将版权转让给提供商等）。
Cloud providers either charge recurring fees or utilise user data for their own profit. This can even pose a security threat to users in unsafe environments (such as journalists or activists) and is not suitable for sensitive or business data (possible infringement of intellectual property, transfer of copyright to the provider, etc.).
此外，在线服务可能随时转变或关闭，例如 Athens Research (2022)，即使可能，数据也不总是能够以有用的方式导出——分类或连接可能会丢失，并且用户无法以与以前相同的方式处理他们的数据（例如，关于搜索、导航或数据丰富）。
Also, online services may pivot or shut down any time, e.g., Athens Research (2022), and even if possible, data cannot always be exported in useful ways – classification or connections may be lost, and users cannot work with their data in the same way as before (e.g., regarding search, navigation, or data enrichment).
最后，大量有价值的信息仍然以精心整理的文档、电子书、论文或其他媒体的形式本地存储在个人桌面系统上，这些文档、电子书、论文或其他媒体可能是受限制的或来自私人来源。这包括个人作品，如项目笔记、想法、概念、各种通信（邮件）、联系人（合作者、外部合作伙伴、作者）和事件（会议、研讨会、会议等），这些作品已经存在于用户的环境中，并且与个人数据和由此产生的工作高度相关。
In the end, a lot of valuable information is still locally stored on personal desktop systems in the form of carefully curated collections of documents, e-books, papers or other media, possibly restricted or from private sources. This includes personal artefacts like project notes, ideas, concepts, various communication (mails), contacts (collaborators, external partners, authors) and events (conferences, workshops, meetings, etc.) already present in the user’s environment, and highly connected to personal data and work that derives from it.
个人知识图谱应原生涵盖所有类型的信息，而不是强制将所有内容都记录在笔记中。最新一波知识管理系统（如 Concept（capacities.io，2022 年）或 Anytype（Anytype Inc.，2023 年））采用了这一方面，称为“类型化”或“基于对象”的笔记记录，认为：
Personal knowledge graphs should cover all kinds of information natively, not forcing everything into notes. This aspect has been adopted by the latest wave of knowledge management systems like Concept (capacities.io, 2022) or Anytype (Anytype Inc., 2023), termed “typed” or “object-based” note-taking, arguing that:
您的整个第二大脑都基于具有类型和不同属性的对象。您不再从空白页开始，而是拥有一个可帮助您捕获信息的结构。
Your whole second brain is based on objects having a type and different properties. You don’t start with an empty page anymore but have a structure that helps you capture information.
（https://capacities.io/）[2]
(https://capacities.io/) [2]
采用与用户环境分离的独立知识管理系统会导致媒体中断，并导致不必要的努力来迁移或复制用户熟悉环境中已经存在的信息，许多知识工作者 (Baty, 2022) 已经观察到了这一点，他们试图完全采用现代 PKG 系统，如 Tana (Tana - The Everything OS, 2022) 或甚至基于桌面的 Obsidian (Obsidian, 2023)。
Adopting a separate knowledge management system, decoupled from the user’s environment, leads to a media break and causes unnecessary effort to migrate or duplicate information already present in the user’s familiar environment, which has been observed by several knowledge workers (Baty, 2022) that tried to fully adopt modern PKG systems like Tana (Tana – The Everything OS, 2022) or even the desktop-based Obsidian (Obsidian, 2023).
因此，真正个性化的知识图谱需要存在于“边缘”，存在于用户的（文件）系统内部。它需要包含现有信息，了解其联系，并使它们可见、可导航和可搜索。
Hence, a truly personal knowledge graph needs to live “on the edge”, inside the user’s (file)system. It needs to embrace existing information, understand its connections, and make them visible, navigable and searchable.
回顾上面提到的 Memex 系统，这与布什的设想非常接近：
Circling back to the Memex system mentioned above, this is quite close to what Bush envisioned:
设想一下未来的个人使用设备，这是一种机械化的私人文件和图书馆……memex 是一种个人存储所有书籍、记录和通信的设备，它是机械化的，因此可以以极快的速度和灵活性进行查阅。它是对他记忆的扩大和亲密补充。
Consider a future device for individual use, which is a sort of mechanized private file and library … A memex is a device in which an individual stores all his books, records, and communications, and which is mechanized so that it may be consulted with exceeding speed and flexibility. It is an enlarged intimate supplement to his memory.
这样的系统需要以数据为中心，并采用标准化、开放的格式来容纳所有类型的内容。它应该提供一个 API，以便以有意义的方式访问数据。这样，数据就摆脱了存储格式的束缚，并作为通用信息公开给应用程序使用。这种方法得到了“以数据为中心的宣言”（Hollister，2022 年）的倡导，并指出：
Such a system needs to put data at the centre and embrace standardised, open formats for all types of content. It should provide an API for accessing data in meaningful ways. Thus, data is freed from storage formats and exposed as universal information for applications to work with. This approach is championed by the “Data-Centric Manifesto” (Hollister, 2022), noting that:
如今，数据被困在企业应用程序和网络平台中。[…] 解决办法是将其颠倒过来。数据是宇宙的中心；应用程序是短暂的。
Data today is trapped in enterprise applications and web platforms. […] The remedy is to flip this on its head. Data is the center of the universe; applications are ephemeral.
该宣言列出了几个也适用于桌面系统的核心原则，包括使用开放标准、自描述数据以及让应用程序充当“访问者”，并得出结论：
The manifesto lists several core principles that also apply to desktop systems, including the use of open standards, self-describing data, and having applications act as “visitors”, concluding that:
数据格局变成了一个洋葱状的结构，其中包含原始或半处理数据（可以想象为数据湖或三重存储库）、本体（可以想象为高级数据模型）来解释数据，以及保护（授权使用并防止数据发生不良情况的层）。
The data landscape becomes an onion-like structure with raw or semi-processed data (think of a data lake or a triple store) and an ontology (think high-level data model) to interpret it and a guard (a layer that authorizes use and prevents bad things from happening to the data).
看看常见的桌面用例，电子邮件不需要重量级的邮件应用程序，而是可以通过本地基础设施服务与提供商同步，并以文件的形式向用户和应用程序公开。用户可以像处理普通文件一样处理电子邮件（搜索提取的元数据、在文件夹之间移动和删除邮件），并在需要时仍选择专门的应用程序进行高级邮件处理。联系人、日历事件或播客也可以这样做。我们将在后面的“BeOS”和“Haiku”部分中看到，这已经在桌面操作系统中实现，并且现在就可以使用了。
Looking at a common desktop use case, instead of requiring heavyweight mail applications, emails could be synchronised with providers by a local infrastructure service and exposed to users and applications alike as files. Users can handle their emails like normal files (search extracted metadata, move between folders and delete mail) and still choose specialised applications for advanced mail handling when needed. Same could be done for contacts, calendar events, or podcasts. We will see later in the sections on “BeOS” and “Haiku”, how this already has been realised in a desktop OS and is available now.
SEN 通过提供语义基础设施来采用并扩展这一概念，使用户能够以透明的方式处理已有数据。信息管理工具不应导致更多的信息过载，而应留在后台以尽量减少精神或技术负担，并清除用户的视图以显示和处理感兴趣的信息。
SEN adopts and extends this concept by providing a semantic infrastructure, allowing users to work with already available data in a transparent way. Information management tools should not lead to even more information overload, but stay in the background to minimise mental or technical overhead, and clear the user’s view to reveal and work with information of interest.
表 6.1 对数据和应用之间关系的不同观点进行了分类。
The table 6.1 categorises different perspectives on the relation between data and applications.
表 6.1
Table 6.1
接下来的部分将说明如何实现这种以数据为中心、以用户为中心的开放信息空间生态系统。建议的解决方案建立在现有桌面操作系统的基础上，并为文件系统和桌面添加了语义功能。
The next sections will illustrate how such an ecosystem of a data-centric and user-focused open information space can be realised. The proposed solution builds on the foundation of an existing desktop operating system and adds semantic functionality to the file system and desktop.
基于语义桌面的个人知识图谱愿景
A Vision for a Personal Knowledge Graph based on the Semantic Desktop
明智地安排你的工作空间 – 为什么个人知识图谱应该存在于你的个人桌面中
Wise up your Workspace – Why a Personal Knowledge Graph should live in your personal desktop
在上一节中，我们确定了需要一种将用户数据放在首位并将应用程序视为以标准化方式处理数据的客户端的环境。将数据与应用程序分离是从现有用户数据构建个人知识图谱的重要要求，因此信息可以独立于最初生成信息的应用程序而获得。
In the previous section, we have established the need for an environment that puts the user’s data first and treats applications as clients that work with it in a standardised way. Decoupling data from applications is an important requirement for building a personal knowledge graph out of existing user data, so information is available independently of the application that originally produced it.
如果我们进一步研究，我们将面临另一个挑战：知识图谱的节点可能代表任何语义实体（书籍、位置、电影、人物），而这不一定对应于常见主流操作系统通常支持的文件类型。如果我们想在将数据导入知识图谱时避免映射或重复数据，我们需要挑战和克服当前桌面操作系统实现中处理文件的限制，并使其适应我们的使用场景。
If we take this further, we face another challenge: nodes of a knowledge graph may represent any semantic entity (Book, Location, Movie, Person), and this does not necessarily correspond to file types usually supported in common mainstream operating systems. If we want to avoid mapping or duplicating data when importing it into a knowledge graph, we need to challenge and overcome limits in current desktop OS implementations for working with files and adapt it to our usage scenario.
文件作为具有元数据属性的实体
Files as Entities with Metadata Attributes
我们可以说文件保存信息，其内容由文件类型定义。就像给定类型的语义实体一样。在语义学中，有一句口头禅，“事物，而不是字符串”，由谷歌 (Singhal, 2012) 推广。知识图谱中的对象应该有意义，而不仅仅是一个标签。在我们的例子中，类似的口头禅是“实体，而不是文件”。
We can say that files hold information, and their content is defined by the file type. Just like semantic entities of a given kind. In semantics, there is a mantra, “things, not strings”, made popular by Google (Singhal, 2012). Objects in a knowledge graph should have meaning, not just a label. In our case, an analogous mantra would be “entities, not files”.
由于实体具有属性，因此需要以可访问的方式将这些属性与源一起存储。我们需要确保元数据在我们的环境中被视为一等公民，并且用户和应用程序都可以轻松发现和操作它。虽然在不同的背景下，但这类似于科学界为提供用于研究的数据的开放访问而引入的 FAIR 原则（Wilkinson 等人，2016 年）。当前主流操作系统大多仅将文件系统元数据用于内部目的，例如，用于文件大小和更改日期等统计信息，或用于访问控制。它们没有为用户提供处理自定义元数据的通用解决方案，而只提供标签或评级等有限的功能。
Since entities have properties, these need to be stored in an accessible way together with the source. We need to ensure that metadata is treated as a first-class citizen in our environment, and that it can be easily discovered and manipulated by users and applications alike. While in a different context, this resembles the FAIR principle introduced in the scientific community for providing open access to data used for research (Wilkinson et al., 2016). Current mainstream operating systems mostly use filesystem metadata for internal purposes only, e.g., for statistical information like file size and change date, or for access control. They do not provide a general solution for users to handle custom metadata, but only offer limited features like labels or ratings.
通过更通用的方法，我们基本上可以将任何类型的信息存储为文件，并将属性存储为元数据，从而允许用户存储任何现实世界或虚拟对象（想想“数字孪生”）、概念或标签的数字表示。文件内容以标准化格式保存数据，并且可以从元数据访问属性，所有这些都通过使用标准文件系统 API 来实现。
With a more general-purpose approach, we can basically store any kind of information as a file, together with properties as metadata, allowing users to store digital representations of any real world or virtual object (think “digital twin”), concept or tag. The file content holds data in a standardised format, and properties are accessible from metadata, all by using the standard filesystem API.
结合关系（见下文），这为我们的环境开辟了广泛的用例，不仅适用于个人知识图谱，还适用于库存等简单用例（书籍或电影库、食谱集合、带有联系人注释的个人 CRM），并减少了对专门应用程序来处理常见日常需求的需求。
Together with relations (see below), this opens up our environment to a vast array of use cases, not only for personal knowledge graphs, but also for simple use cases like inventory (book or movie libraries, recipe collections, a personal CRM with notes on contacts), and reduces the need for specialised applications to handle common, everyday requirements.
这种抽象乍一看可能不寻常，但最近的一个例子（Alexander Obenauer，2021）提出了一种具有相关概念的“实体优先”操作系统：
This abstraction may seem unusual at first, but a more recent example (Alexander Obenauer, 2021) proposes an “entity-first” operating system with related concepts:
在 Graph OS 中，您的所有事物都以节点或项目的形式存在于您的系统中，位于您的图表中。电子邮件、日历事件、文章、网页、播客片段、待办事项列表以及其中的待办事项；一切。并且每件事都可能引用其他事物，或被其他事物引用。
In the Graph OS, all of your things are within your system as nodes, or items, within your graph. Emails, calendar events, articles, web pages, podcast episodes, to do lists as well as the to dos inside them; everything. And each thing may have references to, or be referenced by, any other thing.
实现这种系统的最自然和可行的方法是利用提供语义文件系统的现代操作系统，并内置属性和关系，但目前没有可用的此类操作系统，下一节将概述明显的例外（这也不是我们目的的完整解决方案，但已经足够接近了）。
The most natural and feasible way to implement such a system would be to utilise a modern OS that provides a semantic file system, with attributes and relations built-in, but currently there is no such OS available, with the notable exception outlined in the next section (which is also not a full solution for our purpose, but close enough).
尽管许多现代文件系统现在都支持将自定义属性作为键/值对（通常称为“xattrs”，即“扩展属性”），例如 btrfs、XFS 或 ZFS，但它们仍然不允许查询它们。没有图形化的最终用户支持，[3] 甚至针对高级用户的命令行工具（例如 UNIX find）也不支持扩展属性中的元数据查询，因为它们没有索引，性能会太慢。[4]
Although many modern file systems now support custom attributes as key/value pairs (often called “xattrs” for “extended attributes”), like btrfs, XFS or ZFS, they still don’t allow querying them. There is no graphical end-user support, [3] and even command-line tools such as UNIX find, aimed at advanced users, lack support for metadata queries in extended attributes, since they are not indexed, and performance would be too slow. [4]
最后，最重要的是，任何现代桌面系统都不支持语义关系，从而将桌面语义的使用减少为信息面板中显示的静态元数据（如常见文件浏览器中的“详细信息”视图以显示图像或音频元数据）。
Lastly and most importantly, there is no support for semantic relations in any modern desktop system, reducing the use of desktop semantics to static metadata for display in info panels (like a “details” view in common file browsers to show image or audio metadata).
当前的语义桌面解决方案没有直接在文件系统层实现查询支持和对简单符号链接以外的语义关系的本机支持，而是引入了单独的数据存储，例如嵌入式 SQL 数据库，这增加了大量开销并引入了潜在的数据同步问题。它们通常会添加更符合语义网（W3C，2015）而非桌面的复杂 API，这使开发变得比需要的更困难，并减缓了采用和用户接受度。
Instead of implementing query support and native support for semantic relations beyond simple symbolic links directly in the file system layer, current semantic desktop solutions introduce a separate data storage, such as embedded SQL databases, adding a lot of overhead and introducing potential data synchronisation issues. They often add complex APIs that are more aligned to the semantic web (W3C, 2015) than the desktop, which makes development harder than needed and slows down adoption and user acceptance.
由于类似的复杂性，文件系统开发中更具雄心的努力失败了，试图将成熟的数据库集成到日常使用的桌面操作系统中（参见下面的“历史”部分）。
More ambitious efforts in file system development failed because of similar complexity, trying to integrate a full-fledged database into a desktop OS intended for everyday use (see “History” section below).
理想的解决方案应在通过语义数据链接概念增加价值与底层操作系统所需的更改之间保持谨慎的平衡。该解决方案应该是轻量级的，并且透明自然地与用户熟悉和日常操作的桌面集成，建立在支持语义查询或可以以最小开销扩展的文件系统上。最后，应尽可能重复使用已建立的和众所周知的桌面隐喻（文件夹和文件、拖放、窗口和上下文操作），并且仅在需要时直观地扩展。
An ideal solution should maintain a careful balance between adding value through semantic data linking concepts and the changes needed to the underlying operating system. The solution should be lightweight and integrate transparently and naturally with the desktop the user knows and operates daily, built on a file system that supports semantic queries or can be extended with minimum overhead. Lastly, established and well-known desktop metaphors (folders and files, drag and drop, windows and contextual actions) should be reused as much as possible, and only extended intuitively where needed.
BeOS – 第一个语义桌面操作系统
BeOS – The First Semantic Desktop OS
现在几乎被遗忘了，但仍然被一群忠实的前用户和开发人员（包括作者）牢牢记住并高度推崇，1995 年 10 月，一家名为 Be Inc. 的不知名公司推出了一款从头编写的全新操作系统，名为 BeOS。
Now almost forgotten, but still well remembered and highly regarded by a loyal group of former users and developers (including the author), in October 1995, an unknown company called Be Inc. introduced a completely new operating system written from scratch called BeOS.
它迅速占领了一个规模虽小但热情高涨的市场，并支持当时其他桌面操作系统完全不存在的新概念。（Pinheiro，2020 年）。
It took a small but enthusiastic niche by storm and supported novel concepts that were completely absent from other desktop operating systems at that time. (Pinheiro, 2020).
对于最终用户来说，文件系统可能是最强大的功能，因为它支持可索引查询的自定义属性，并且使用起来非常方便。即使是标准文件浏览器（称为“Tracker”）也包含文件系统属性，可以在常见的表格式详细视图中显示文件元数据。
The filesystem was probably the most powerful feature for end users, as it supported custom attributes that could be indexed for querying, and a user-friendly approach to work with them. Even the standard file browser (called “Tracker”) embraced file system attributes, showing file metadata in a common table-like detail view.
这样就实现了通用的互操作性（正如以数据为中心的宣言中所宣称的那样），并使用户更容易处理他们的数据，开发人员也可以专注于标准操作系统或桌面功能尚未支持的高级用例。[5]
This allowed for universal interoperability (as proclaimed in the data-centric manifesto) and made it easier both for users to work with their data, and for developers who could focus on advanced use cases not already supported by standard OS or desktop features. [5]
此外，联系人管理 [6] 或电子邮件 [7] 等标准功能也是使用文件系统元数据实现的，这些元数据以开放、可访问的格式存储信息。然后使用文件系统查询向用户呈现相关内容，例如显示新邮件或筛选发件人或主题。类似的概念适用于联系人和音频文件，这些文件的嵌入元数据（如 ID3 标签）被提取到元数据属性中。
Also, standard functionality like contact management [6] or email [7] was implemented using file system metadata for storing information in an open, accessible format. Filesystem queries were then used to present relevant content to the user, like showing new mail or filtering for sender or subject. Similar concepts applied to contacts and audio files, which had their embedded metadata (like ID3 tags) extracted into metadata attributes.
图 6.1 表示可以保存联系人或任何其他用户数据的实体的文件，其属性存储在 BeOS 的文件系统属性中。
Figure 6.1 Files representing entities that can hold contacts or any other user data with properties stored in filesystem attributes in BeOS.
然而，用户无需使用专门的应用程序就可以直接操作数据，因为元数据在文件浏览器中随时可用，并且内容以标准数据格式存储，如电子邮件的 RFC 标准。因此，现实世界的实体直观地表示为桌面上的文件，并且它们的属性对用户公开可用。
Users could however just operate on their data without a specialised application, since metadata was readily available in the file browser, and the content was stored in standard data formats, like RFC standards for email. So, entities of the real world were intuitively represented as files on the desktop, and their properties were openly available to users.
有了这些设计决策，对专用应用程序的需求就会减少，因为许多常见用例已经通过扩展用户从其他系统了解的常用桌面隐喻来支持。通过将丰富的元数据支持作为文件系统的核心功能，并在文件浏览器中提供可视化支持，最终用户可以实现简单的以数据为中心（“CRUD”）用例，例如用于库存管理（书籍、记录或食谱），以及更高级的用例，而无需开发或使用专用应用程序。[8]
With these design decisions, the need for specialised applications could be reduced, as many common use cases were already supported by extending commonly established desktop metaphors users already knew from other systems. By implementing rich metadata support as a core feature of the filesystem with visual support in the file browser, simple data-centric (“CRUD”) use cases, e.g., for inventory management (books, records or recipes), but also more advanced use cases could be realised by end users, without the need to develop or use dedicated applications. [8]
该项目网站 (Humdinger, 2019) 提供的研讨会也说明了这一点，如图 6.2 所示。
This is also illustrated in a workshop provided on the project website (Humdinger, 2019), as shown in figure 6.2.
图 6.2 在 BeOS 中查询知识图谱中的自定义文件类型（如实体）
Figure 6.2 Querying for custom file types like entities in a knowledge graph in BeOS
文件类型基于 MIME 类型（IANA，2022），实体的属性存储在自定义文件系统属性中，只有关系被省略，因为 BeOS 的原始创建者发现了与上一节概述的相同的问题；尽管该操作系统的第一个版本仍然有一个表和关系 API。
File types were based on MIME types (IANA, 2022), properties of entities were stored in custom file system attributes, only relations have been omitted because the original creators of BeOS identified the same issues as outlined in the previous section; although the first version of the OS still had a Table and Relations API.
所有这些都已经非常接近之前提到的“GraphOS”，用户可以对存储在文件系统中的实体和元数据进行操作，并且早在 1995 年就已面世。
All of this already comes pretty close to the “GraphOS” mentioned before, where users operate on entities and metadata stored in the filesystem, and was already available in 1995.
尽管 BeOS 在技术上很先进，但仍然以用户为导向，为日常用例提供了创新的解决方案，但它也在许多方面标志着一种明显的范式转变，并给开发人员带来了新的挑战。此外，当时桌面市场由微软主导，而 Be Inc. 作为一家不知名的小公司，无法获得足够的吸引力来生存。幸运的是，具有上述概念的现代桌面操作系统的愿景在一群开源开发人员中引起了关注，因此 Haiku（Haiku, Inc.，2022 年）诞生了。但在介绍 Haiku 作为 SEN 的基础之前，让我们简要回顾一下试图将语义引入桌面的并行开发，并从他们的错误中吸取教训。
Although BeOS was technically advanced but still user oriented, providing innovative solutions for everyday use cases, it also marked a stark paradigm shift in many ways and posed new challenges to developers. Also, the desktop market was dominated by Microsoft back then, and Be Inc. as a small unknown company could not gain enough traction to survive. Fortunately, the vision of a modern desktop operating system with the concepts introduced above found traction among a group of open source developers, and so Haiku (Haiku, Inc., 2022) was born. But before introducing Haiku as the basis for SEN, let’s have a short look back at parallel developments that tried to bring semantics to the desktop, and learn from their mistakes.
从过去吸取教训：语义桌面的前景和失败简史
Learning from the Past: A short history of promises and failures of the Semantic Desktop
Linux KDE 桌面引入了语义桌面和语义应用程序的创新概念（Sauermann 等人，2009），包括简单的扩展和混搭，丰富了现有应用程序（SEN 也持这种概念）。它作为 KDE Linux 桌面的一部分，以 NEPOMUK（Bernardi 等人，2011）的形式实现，这是一个雄心勃勃的欧盟资助研究项目。
The Linux KDE desktop introduced an innovative concept for a semantic desktop and semantic applications (Sauermann et al., 2009), including just simple extensions and mashups, enriching existing applications (a concept shared by SEN). It was realised as part of the KDE Linux desktop in the form of NEPOMUK (Bernardi et al., 2011), an ambitious EU funded research project.
然而，在该项目的整个生命周期中，使用功能齐全的数据库来存储语义元数据很快就出现了问题。此外，像 RDF 这样的复杂标准被用作处理文件类型和元数据的原生格式，这使得即使是简单的任务也变得成本高昂，而且不太方便用户使用。因此，用户遇到了许多性能下降的问题，很快 NEPOMUK 及其索引器就被认定为主要的性能消耗者。
Over the course of the project’s lifetime, however, it quickly became problematic to use a full-fledged database to store semantic metadata. Also, complex standards like RDF were used as native format for handling file types and metadata, making even simple tasks expensive and not very user friendly. As a result, users had many issues with degrading performance, and soon NEPOMUK and its indexer were identified as a major performance hog.
最新版本采用了更轻量、更灵活的架构，使用名为“Baloo”的文件索引和搜索框架（Baloo – KDE 社区维基，2022 年），“专注于提供非常小的内存占用和极快的搜索速度”，但即便如此，也会导致严重的性能影响和存储开销。[9]
The latest incarnation follows a more lightweight and flexible architecture, using a file indexing and search framework called “Baloo” (Baloo – KDE Community Wiki, 2022) “with a focus on providing a very small memory footprint along with extremely fast searching”, but even that can cause a severe performance impact and storage overhead. [9]
主流操作系统也尝试过类似的方法但失败了，例如微软的 WinFS（LSoft Technologies Inc.，2022）——另请参见（Orlowski，2002）中由更成功且实际上完全实现的 BeOS 文件系统 BFS 的开发人员所做的批判性但富有洞察力的讨论，该讨论将在下一节中详细介绍。
Similar approaches were also tried in mainstream operating systems but failed, e.g., Microsoft’s WinFS (LSoft Technologies Inc., 2022) – see also a critical but insightful discussion in (Orlowski, 2002) by the developers of the more successful and, actually, fully realised BeOS File System BFS, which is detailed in the next section.
当前语义桌面解决方案的问题大致可以概括如下：
Problems with current semantic desktop solutions can be roughly summarised as follows:
陷入复杂性陷阱：将全套语义网标准应用于桌面，尽管其非常复杂，但事实证明并不可行。例如，Gnowsys (Nagarjuna，2013) 最初是一个雄心勃勃的项目，目标是构建“语义计算内核”，但自 2013 年以来一直停滞不前。
Falling into the Complexity Trap: Applying the full set of semantic web standards to the desktop with all its complexity proves not really feasible. E.g., Gnowsys (Nagarjuna, 2013) started as an ambitious project with the goal to build a “A Kernel for Semantic Computing”, but has been stale since 2013.
漏洞百出的抽象：暴露了语义（网络）标准的全部范围，例如用于存储的 RDF 或用于搜索的 SPARQL，同时忽略了用户体验，吓跑了许多潜在用户，并将解决方案限制为仅供专家用户使用。这阻碍了广泛采用，进而阻碍了开发人员的支持，而开发人员的支持是扩展和集成应用程序所必需的。
Leaky Abstractions: exposing the full breadth of semantic (web) standards like RDF for storage, or SPARQL for search, while neglecting the user experience, scares off many potential users and restricts the solution to expert users only. This hinders widespread adoption and in turn developer support, which is needed for extending and integrating applications.
性能影响：当索引和元数据提取造成的资源影响过高时，其成本会超过语义桌面功能带来的好处。
Performance Impact: when resource impact caused by indexing and metadata extraction is too high, the cost outweighs the benefits from semantic desktop features.
缺少查询和导航支持：应该以简单直观的方式支持超越简单元数据提取的语义链接数据的强大导航和搜索。
Missing Query and Navigation Support: Powerful navigation and search on semantically linked data beyond simple metadata extraction should be supported in a simple and intuitive way.
Haiku – 完美的原型设计环境
Haiku – the Perfect Prototyping Environment
Haiku 最初名为“OpenBeOS”，其愿景是将 BeOS 重建为一个开源项目，坚持其仍然富有远见的设计目标，但谨慎地调整操作系统和桌面以符合当前标准。这包括现代软件包管理器、Web 堆栈、主要开源库和应用程序的端口以及现代驱动程序支持。
Originally called “OpenBeOS”, Haiku’s vision is to rebuild BeOS as an open source project, adhering to its still visionary design goals, but carefully adapting the OS and desktop to current standards. This includes a modern package manager, web stack, ports of major open source libraries and applications, and modern driver support.
在 2017 年开源 FOSDEM 大会 (Revol, 2017) 上举行的一次演讲中，可以找到对 Haiku 及其仍然具有创新性的概念的良好介绍。2007 年的一次 Google 技术演讲 (Google, 2007) 详细介绍了其历史和发展，演讲嘉宾是 Be Inc.（BeOS 背后的公司）的创始人兼前首席执行官 Jean-Louis Gassée。
A good introduction to Haiku and its still innovative concepts can be found in a presentation held at the open source FOSDEM conference in 2017 (Revol, 2017). A detailed look into its history and development is provided in a Google tech talk of 2007 (Google, 2007), starring Jean-Louis Gassée, the founder and ex-CEO of Be Inc., the company behind BeOS.
那么，尽管 Haiku 的市场份额很小，但它为什么能够成为实现作为个人知识图谱基础的语义桌面的良好候选者呢？
So what makes Haiku a good candidate, despite its niche market share, for realising a semantic desktop that acts as the foundation for a personal knowledge graph?
Haiku 为终端用户提供了一个低调、以桌面为中心的工作环境。该操作系统采用了熟悉的概念，但在许多方面进行了扩展，从创新的技术概念到用户界面设计功能。这使得它成为语义桌面用例和知识工作者的理想选择。
Haiku provides a low-profile, working desktop-centric environment targeting end users. The OS embraces familiar concepts but extends them in many ways, from innovative technical concepts to user-interface design features. This makes it an ideal fit for semantic desktop use cases and knowledge workers.
Haiku 提供了像 SEN 这样的系统所需的关键概念：具有本机元数据支持的文件系统、允许对保存元数据的自定义文件系统属性进行快速、基于索引的查询的丰富 API，以及展示这些概念的易于使用的桌面界面。
Haiku provides key concepts needed for a system like SEN out of the box: a filesystem with native metadata support, a rich API that allows fast, index-based querying on the custom filesystem attributes holding metadata, and an easy-to-use desktop interface exposing these concepts.
此外，应用程序之间广泛使用消息传递，可以无缝集成 SEN 提供的语义功能。只要支持应用程序提供足够的消息来处理相应的操作，用户就可以在文档或媒体文件内导航关系并跳转到关系目标。[10]
Also, the pervasive use of message passing between applications allows for seamless integration with semantic functionality provided by SEN. Users can navigate relations and jump to relation targets inside documents or media files, as long as the supporting application provides adequate messages to handle respective actions. [10]
对用户数据的整体视图（其中文件可以容纳任何类型的实体）最能反映用户的世界并将其对象映射到本机文件系统。然后，用户可以像处理任何其他系统上的文件一样创建和操作这些实体。这弥补了常见的桌面使用和文档操作与个人知识图谱中涉及语义关系的更高级用例之间的差距——实际上，对于 SEN，文件系统就是知识图谱，无需用户使用专门的软件来构建和维护它。
A holistic view on user data, where files can hold any kind of entities, best reflects the user’s world and maps its objects to the native filesystem. Users can then create and manipulate these entities in the same way as they would handle files on any other system. This bridges the gap between common desktop use and document manipulation, and more advanced use cases involving semantic relations in a personal knowledge graph – actually, with SEN, the filesystem is the knowledge graph without requiring users to use specialised software to build and maintain it.
图 6.3 在文件管理器 Tracker 中通过模板列表创建新文件
Figure 6.3 Creating a new file from a list of templates in Tracker, the file manager
这个简单但强大的概念使知识图谱始终与用户收集的信息保持同步，并避免任何断开、重复或冲突。
This simple but powerful concept keeps the knowledge graph always in sync with the information collected by users and avoids any disconnect, duplication or conflicts.
虽然从理论上讲，这些概念可以应用于更主流的操作系统，如 Linux 甚至 MacOS，但最终的解决方案会让人感觉不自然和不合适，因为这些系统从来就不是为这种方法设计的。
While in theory, these concepts could be applied to more mainstream operating systems like Linux or even MacOS, the resulting solution would feel unnatural and out of place, since these systems were never designed for such an approach.
因此，考虑到之前列出的要求和动机，Haiku 为个人知识图谱提供了一个可行的平台。即使它是一个小众操作系统，开源和极简主义方法与支持语义用例的强大概念相结合，也使其成为知识工作者，甚至是对新颖的笔记解决方案感兴趣的普通用户的完美选择。由于 Haiku 可以通过 USB 记忆棒或虚拟机运行，因此它非常适合专注、专注的知识工作，例如“禅模式”，这也是项目名称“SEN”的双关语。
As a result, Haiku makes a viable platform for a personal knowledge graph given the requirements and motivation laid out before. Even if it is a niche operating system, the open source and minimalist approach combined with powerful concepts to support semantic use cases make it a perfect fit for knowledge workers, or even average users that are curious about novel note-taking solutions. Since Haiku can be run from a USB stick or a virtual machine, it is very well suited for focused, dedicated knowledge work, like a “Zen mode”, which is also an intended pun of the project name “SEN”.
介绍 SEN – 一种现代简约的以用户为中心的方法
Introducing SEN – a modern minimalist user-centric approach
SEN（Rosenauer，2023）通过利用和扩展 Haiku 已经提供的丰富基础设施和 API 实现了前面几节中概述的环境：文件自然代表实体，因为类型系统基于 MIME 类型。实体属性存储在自定义文件系统属性中；仅缺少关系。
SEN (Rosenauer, 2023) realises the environment outlined in previous sections by utilising and extending the rich infrastructure and API already provided by Haiku: files naturally represent entities, as the type system is based on MIME types. Entity properties are stored in custom filesystem attributes; only relations are missing.
SEN 通过将关系建模为实体属性并将其与其他属性一起存储为文件系统属性，为完全基于文件系统的语义桌面添加了最后一个缺失的部分。最重要的是，SEN 提供了一个精简的基于消息的 API，以便在需要和可行的情况下将这些语义扩展集成到桌面和现有应用程序中。
SEN adds this last missing piece for a completely file system based semantic desktop by modelling relations as entity properties and storing them together with other properties as file system attributes. On top, SEN provides a lean, message-based API to integrate these semantic extensions into the desktop and existing applications, where needed and feasible.
为了提高性能，任何属于关系的文件都会获得一个唯一且稳定的标识符（如文档管理系统或数据库中的对象 ID），并且关系会在单个自定义属性中引用此 ID。由于源文件的对象 ID 和目标文件的关系 ID 都存储在索引属性中（称为“SEN:ID”和“SEN:TARGETS”），因此可以非常高效地查询它们。
For performance, any file that is part of a relation gets a unique and stable identifier (like an Object ID in a document management system or database), and relations reference this ID in a single custom attribute. Because both the Object ID of the source file and the relation IDs of the target files are stored in indexed attributes (called “SEN:ID” and “SEN:TARGETS”), they can be queried very efficiently.
关系属性存储在无需索引的附加属性中，因为 SEN 会近乎实时地按需检索它们。它们作为每个关系的键/值对集合存储在单独的属性中，并可提供技术和面向用户的信息。例如，人→书关系可以使用“角色”属性来描述人在该关系中的角色，而“标签”则更为通用，可用于将人→位置关系分类为“工作”和“家庭”地址。有关关系的技术信息可能包括文档中的位置、书中的页面或媒体文件中的时间码。
Relation properties are stored in additional attributes that need not be indexed, as SEN retrieves them on demand in near real time. They are stored in separate attributes as a collection of key/value pairs for each relation and may provide both technical and user-facing information. E.g., a Person→Book relation could describe the person’s role in that relationship by using a “role” attribute, whereas a “label” is more generic and could be used to categorise a Person→Location relation in “work” and “home” address. Technical information about the relation may include a position in a document, page in a book, or timecode in a media file.
设计理念和核心原则
Design Philosophy and Core Principles
简单：SEN 不是专家系统，而是针对个人桌面和普通用户：应根据需要和理解逐步透明地提供更高级的功能
Simple: SEN is not an expert system, but targeted at personal desktop and average users: should gradually and transparently provide more advanced functionality as needed and understood
不引人注目：不会显著影响系统性能和资源
Unobtrusive: should not notably impact system performance and resources
透明：应与桌面和常见隐喻集成，直接处理文件、文件夹、文件类型和属性 - 仅在需要时才扩展到桌面（文件管理器、行为）（例如，集成新的关系概念）
Transparent: should integrate with desktop and common metaphors, working directly on files, folders, file types and attributes – extensions to desktop (file manager, behaviour) only where needed (e.g., to integrate new concept of relations)
开放但保密：系统应开放以供扩展，但保持个人数据的私密性：通过插件进行扩展（例如，从文件中提取属性和实体），导入数据（例如，来自 schema.org 的本体或单个实体），但仅在明确请求时交换数据（用户之间的链接数据）
Open but private: system should be open for extension, but keep personal data private: extension through plugins (e.g., for extracting attributes and entities from files), for importing data (e.g., ontologies or individual entities from schema.org), but exchanging data only when explicitly requested (linked data between users)
拥抱软件工具隐喻：通过构建标准操作系统语义（特别是 POSIX）和先进但标准化的文件系统概念，SEN 可以利用早在 1976 年就已流行的管道和过滤器模式（Kernighan & Plauger，1976）：工具旨在很好地执行单一目的，但也实现标准接口以实现无缝互操作性。这允许将简单的命令链接在一起以执行更复杂的任务（然后使用 UNIX“管道”，但在这种情况下，通过文件属性、脚本和消息传递在更高级别上）
Embrace the software tools metaphor: by building on standard operating system semantics (specifically, POSIX) and advanced but standardised file system concepts, SEN can make use of the pipes&filters pattern first popularised already in 1976 (Kernighan & Plauger, 1976): tools are designed to perform a single purpose well but also implement standard interfaces for seamless interoperability. This allows chaining simple commands together to perform more complex tasks (then using UNIX “pipes”, but in this context, on a higher level through file attributes, scripting and messaging)
基本架构
Basic Architecture
SEN 的核心是作为系统守护进程（在 Haiku 中称为“服务器”）实现的，它处理与文件系统和元数据属性交互的所有基本功能，包括其上的关系。
The core of SEN is implemented as a system daemon (called “server” in Haiku) that handles all basic functionality for interacting with the file system and metadata attributes, including relations on top of it.
核心组件还处理具有明确定义属性的可用关系的配置，并与现有应用程序集成以透明地解析语义链接，例如打开相关文档或联系人。甚至可以实现“深度链接”，因此用户可以在特定位置打开相关文件，或在单独的注释中显示注释引用的突出显示的文本（参见下面的示例）。这仅要求处理目标文件的应用程序支持脚本 [11] 消息命令以进行导航和突出显示，但不需要开发人员调整应用程序并实现特定的 SEN API。
The core component also handles configuration of available relations with well-defined attributes, and integrates with existing applications to transparently resolve semantic links, e.g., opening related documents or contacts. Even “deep linking” is possible, so users can open a related file at a specific position or show a highlighted piece of text referenced by an annotation in a separate note (see examples below). This only requires applications that handle a target file to support scripting [11] message commands for navigation and highlighting, but does not require developers to adapt applications and implement specific SEN APIs.
SEN 不仅扩展了底层系统，而且与现有应用程序原生集成，仅在必要时添加轻微的调整（例如脚本支持）。这为用户和开发人员创建了一个有凝聚力和整体的语义桌面解决方案；该解决方案不会取代现有的桌面环境，而只会用语义丰富它。同样，它不会强迫开发人员采用特殊的 API 来为应用程序添加语义功能。甚至为上述“深层链接”添加（标准）脚本支持也可以看作是独立于 SEN 的一般功能，为应用程序增加了价值。
SEN not only extends the underlying system, but integrates natively with existing applications, only adding slight adaptations (e.g., scripting support) where necessary. This creates a cohesive and holistic semantic desktop solution for users and developers; this solution does not replace an existing desktop environment, but only enriches it with semantics. Similarly, it does not force developers to adopt special APIs for adding semantic functionality to applications. Even adding (standard) scripting support for aforementioned “deep links” can be seen as a general feature adding value to the application, independently of SEN.
图 6.4 简要概述了技术解决方案以及 SEN 的适用范围。
Figure 6.4 gives a quick overview of the technical solution and where SEN fits in.
图 6.4. SEN 技术架构概览
Figure 6.4. Overview of the technical architecture of SEN
如图所示，SEN 利用操作系统提供的文件系统和存储 API，仅在标准 OS API 和应用程序之间添加一个轻量级语义扩展层。SEN 由各种组件组成，这些组件通过消息传递（使用 Haiku 的标准消息传递 API）进行交互，既在内部进行，也与文件浏览器“Tracker”等应用程序进行交互。这产生了一种松散耦合的架构，可以与现有工具集成，提供从文件中提取元数据以供 PKG 使用等功能，如下所述。
As shown in the diagram, SEN makes use of the filesystem and storage API provided by the operating system, adding only a lightweight semantic extension layer between standard OS APIs and applications. SEN consists of various components that interact via messaging (using the standard messaging API of Haiku), both internally and with applications like the file browser “Tracker”. This results in a loosely coupled architecture that can integrate with existing tools providing functionality like metadata extraction from files for use in a PKG, as described below.
整合信息提取与文档分析
Integrating Information Extraction and Document Analysis
上面概述的架构允许与语义和文档分析领域中众所周知和广受欢迎的大量现有、成熟和开源工具进行强大且非常灵活的集成。
The architecture outlined above allows for a powerful and very flexible integration with a vast array of existing, mature and open source tools well known and widely popular in the semantic and document analysis domain.
由于 Python 在 Haiku 上开箱即用，SEN 只需要包装和调整脚本，以便将结果作为元数据存储在文件系统属性中，然后就可以在系统范围、以数据为中心的方式重复使用它们 (Hollister，2022)。
Since Python runs out of the box on Haiku, SEN only needs to wrap and adapt scripts so they store results as metadata in file system attributes, where they can be reused in a system-wide, data-centric manner (Hollister, 2022).
所有流行的文档分析工具都可轻松用于从各种格式的文档中提取信息。它们可通过脚本组合起来，以 SEN 所需的形式（即定义明确的文件属性）存储提取的信息。一旦信息以这种规范化形式提供，其他应用程序便可独立于 SEN 透明地访问它。
All the popular document analysis tools can be readily used to extract information from documents in various formats. They can be combined through scripting for storing extracted information in the form required by SEN, namely well-defined file attributes. Once the information is available in this normalised form, it can be transparently accessed by other applications independently of SEN.
此类工具可用于简单的元数据提取（例如，文档中的作者、关键字、页数，或 EXIF/IPTC 信息中的图像元数据，或比特率、格式、字幕信息、长度甚至语言等音频/视频元数据）。更高级的工具可以提供完整的文档理解和分类，通过情绪分析提取语义信息，从发票中提取特定信息，引用人物、组织、地点或日期等。
Such tools might be used for simple metadata extraction (like author, keywords, page count from documents, or image metadata from EXIF/IPTC information, or audio/video metadata like bitrate, format, subtitle information, length or even language). More advanced tools may provide full document understanding and classification, extracting semantic information through sentiment analysis, specific information from invoices, references to people, organisations, locations or dates, etc.
高级解决方案可以将几种工具组合成工具链，并使用强大的文档分类系统（如 Donut（Kim et al.，2022））提供集成解决方案（如（DocQuery，2022/2022），也可以独立使用。
Advanced solutions may combine several tools into tool chains and provide integrated solutions like (DocQuery, 2022/2022) using powerful document classification systems like Donut (Kim et al., 2022), which can also be used stand-alone.
关系——缺失的环节
Relations – The missing link
关系是所提解决方案的基石，将文件之间的关系以及关系属性存储在文件系统属性中。这使得语义桌面可以作为个人知识图谱的基础，如愿景声明中所述。
Relations are the cornerstone of the proposed solution, storing relations between files, along with relation properties, in filesystem attributes. This enables a semantic desktop as the basis for a personal knowledge graph, as described in the vision statement.
因此，与 UNIX 标准“符号链接”（通常使用“ln”命令创建的文件之间的符号链接）不同，SEN 引入了语义链接，它依赖于自定义属性中的文件系统元数据，但将信息存储在文件系统中。这可以防止磁盘上的信息和知识图中的信息之间出现任何脱节，并提供与用户环境完美集成的本机集成，而无需额外的数据库或专用工具。
So as opposed to UNIX standard “symlinks” (symbolic links between files usually created with the “ln” command), SEN introduces semantic links, which rely on filesystem metadata in custom attributes, but keep the information stored in the filesystem. This prevents any disconnect between information on disk and in the knowledge graph, and provides a native integration well integrated into the user environment, without additional requirements for a database or specialised tools.
SEN 提供了一个 API，可根据源文件类型和关系配置筛选合适的关系。应用程序使用此 API 为用户提供一种在文件上指定有意义关系的方法。需要强调的是，SEN 完全支持以数据为中心的方法，提供该 API 主要是为了方便使用和保持系统一致性。但是，没有锁定，应用程序可以直接访问相关的文件系统属性。
SEN provides an API to filter suitable relations based on the source file type and relation configuration. Applications use this API to provide a way for users to specify meaningful relations on files. It is important to stress that SEN fully supports a data-centric approach, and the API is provided mostly for convenience and to keep the system consistent. However, there is no lock-in, and applications can access the relevant file system attributes directly.
例如，文件浏览器 Tracker 的增强版本使用 SEN API 在菜单中显示适用于所选文件的关系（“打开相关...”），并在单独的窗口中提供更详细的视图，其中关系属性显示为列，如标准属性“名称”和“类型”。这将在“在文件管理器“Tracker”中可视化和导航关系”部分中更详细地解释。
For example, an enhanced version of the file browser Tracker uses the SEN API to display relations applicable to the selected file(s) in a menu (“Open related…”), and provides a more detailed view in a separate window, where relation properties are shown as columns, like the standard attributes “name” and “type”. This is explained in more detail in section “Visualising and Navigating Relations in the File Manager ‘Tracker’”.
扩展的文本编辑器可以使用 SEN API 查询相关内容和对其他文本文件、人物、地点、概念或媒体来源（书籍、电影或歌曲）的引用。这还包括自我引用，如书籍中的章节（PDF 文档或电子书）、播客中的部分（音频文件）或源代码文件中的方法。结果可以根据需要进行可视化，并适合应用程序，例如树形视图或列表，或内联嵌入到文档中（请参阅“UNO - 通用笔记本的概念”部分中简单的语义记事本示例）。
An extended text editor may use the SEN API to query for related content and references to other text files, people, locations, concepts, or media sources (books, movies or songs). This also included self-references like chapters in books (PDF documents or e-books), sections in a podcast (audio file), or methods in source code files. The result can be visualised as needed and suitable for the application, e.g., a tree view or list, or inline embedded into the document (see the example for a simple semantic notepad in section “UNO – a Concept for a Universal NOtebook”).
三元关系
Ternary Relations
然而，在许多用例中遇到的一种特殊关系类型是对三元关系的支持，即当三个实体是关系的一部分时。
A special relationship type that is nevertheless encountered in many use cases is the support for ternary relations, i.e., when three entities are part of a relation.
一个例子是电影注释中引用出现在特定场景中的演员之间的关系。见图 6.5。
An example would be relations between notes on a movie referencing actors that appear in certain scenes. See figure 6.5.
图 6.5 三元关系
Figure 6.5 Ternary Relations
为了解决这种情况，在仍然与上面建立的关系到文件属性的映射保持一致的同时，我们可以将对其他实体的引用与正常关系属性一起存储，使用保留标签（如 SEN:ID）作为属性键，并使用目标实体的 ID 作为属性值。
To cover this case, while still staying consistent with the mapping of relations to file attributes as established above, we can store references to other entities along with normal relationship properties, using a reserved label like SEN:ID as property key, and the ID of the target entity as property value.
如上所述，我们可以使用属性值中的集合来涵盖对具有不同属性的单个实体有多个引用的特殊情况，例如，不同场景中的同一个演员，或在不同时间码对同一位置的引用。
We can use collections in property values, as outlined above, to cover the special case of having multiple references to a single entity with different properties, e.g., the same actor in different scenes, or references to the same location at different time codes.
查询 – 解决关系
Queries – resolving relations
Haiku 原生支持对自定义文件系统属性的查询，使用简单的自定义查询语言和索引来提高性能（Humdinger，2019）。还有一个简单的查询 UI，允许用户通过搜索字段输入查询并指定各种选项来缩小查询范围，如图 6.2 所示。这允许用户根据存储在这些属性中的元数据快速检索文件。
Haiku natively supports queries on custom filesystem attributes, using a simple, custom query language and an index for faster performance (Humdinger, 2019). There is also a simple Query UI that allows users to enter queries through a search field and specify various options to narrow down the query, as shown in figure 6.2. This allows users to quickly retrieve files based on metadata stored in these attributes.
由于 SEN 还将关系引用存储在属性中，因此可以通过简单地查询具有 SEN:ID 或 SEN:TARGETS 属性的文件（该属性包含关系所需端的给定 SEN:ID）来本地且透明地解析它们。
Because SEN also stores relation references in attributes, they can be natively and transparently resolved by simply querying for files with a SEN:ID or SEN:TARGETS attribute that contains the given SEN:ID of the desired end of a relation.
因此，为了解析 SEN:ID 为 0815 的文件的所有 contributesTo 类型关系，SEN 服务器使用 Haiku 文件系统 API 执行本机文件系统查询，以查找 SEN:TARGETS 属性包含字符串 0815 的所有文件。
So, for resolving all relations of type contributesTo of a file with SEN:ID of 0815, the SEN server executes a native file system query using the Haiku filesystem API to find all files with a SEN:TARGETS attribute containing the String 0815.
为了使查询有效（并确保足够的性能），需要对属性进行索引。此索引存储在磁盘上。
For queries to work (and to ensure adequate performance), attributes need to be indexed. This index is stored on disk.
SEN 仅索引 SEN:ID 和 SEN:TARGETS 属性，而不索引包含关系属性的属性，以免在绝对必要的情况下过度破坏文件系统索引。这意味着查询特定目标（需要子字符串匹配）比查询源的成本更高，但这是一个符合使用模式的有效折衷方案，因为查询 SEN ID 发生在解析引用文件时，这比解析指向给定文件的反向链接更常见。
SEN only indexes the SEN:ID and SEN:TARGETS attributes, but not attributes holding relation properties, to not blow up the filesystem index more than absolutely necessary. This means that querying for specific targets (requiring substring matching) is more costly than querying for sources, but this is a valid compromise that fits the usage pattern, since querying for SEN IDs happens when resolving referenced files, which will be more common than resolving back links leading to a given file.
关系属性仅在需要时使用，例如，解析关系目标中的位置（文本偏移量、视频时间码）或纯粹的描述性质（角色、标签、注释）。因此，它们直接从元数据中获取，无需通过文件系统查询进行解析。
Relation properties are only used on demand, e.g., resolving a position in a relation target (text offset, video timecode), or of purely descriptive nature (role, label, annotation). Hence, they are taken directly from metadata and need not be resolved through file system queries.
正式模式定义和 SEN
Formal Schema Definitions and SEN
正如我们上面所看到的，SEN 需要确保提取的数据以通用的属性名称和格式存储，以便可以以统一的方式在知识图谱中使用关系。[12] 这对于可视化（如突出显示文档中引用的部分）和导航也至关重要。要做到这一点，需要对可用的实体、关系和属性进行一些定义。在语义上，这被称为本体（Berman，2022 年）。
As we have seen above, SEN needs to ensure that extracted data is stored in common attribute names and formats, so relations can be used in the knowledge graph in a uniform way. [12] This is also essential for visualisation (like highlighting referenced sections in documents) and navigation to work. For this to work, there needs to be some definition of available entities, relations and properties. In semantics, this is called an ontology (Berman, 2022).
有意不采用正式模式，以保持系统的实用性、最小化其影响，并使解决方案易于非专家用户使用，因为这些用户可能不熟悉语义概念和格式，如 OWL、RDF 或其变体。SEN 采用的方法（并得到底层 OS 的认可）类似于使用用户控制元数据的 folkonomies（Vander Wal，2007），甚至 W3C 也考虑过这种方法（Schepers，2007）。这降低了进入门槛，同时仍然通过强大但易于使用且更自然的语义功能大大丰富了用户的通用环境。
A formal schema is deliberately not endorsed to keep the system practical, its impact minimal, and the solution approachable for non-expert users, who might not be familiar with semantic concepts and formats like OWL, RDF or its variants. The approach adopted by SEN (and endorsed by the underlying OS) is similar to folksonomies (Vander Wal, 2007) with user-controlled metadata, as even considered by the W3C (Schepers, 2007). This lowers the barrier to entry and still enriches the user’s common environment considerably with semantic features that are powerful but easy and more natural to use.
Schema.org 定义了一组实体和标准属性，Wikidata 和其他来源（如 FOAF）也定义了一组实体和标准属性（FOAF 词汇规范，2014 年）。
Schema.org defines a set of entities and standard properties, as does Wikidata and other sources like FOAF (FOAF Vocabulary Specification, 2014).
举个例子，为了记录我们读过的书籍，我们会定义一个文件类型“书籍”，它具有常见的、众所周知的属性，如 ISBN、作者、标题和关键词。然后，这些信息要么由单独的工具从电子书或 PDF 文件中提取，要么由用户手动输入，并存储在这些定义明确的属性中，以便于普遍访问。从注释到特定书籍的引用将使用标准 startOffset 属性作为关系属性，因此 SEN 可以使用它来导航到链接文档中的指定位置。现在可以像普通文件一样搜索和管理书籍，而无需专门的图书馆应用程序。这些对于导入、导出或转换等高级用例仍然有用，但对于基本用例来说不是必需的。[13] 要求用户配置一组一致的文件类型、属性和关系会违背简单易用的目标。
As an example, for keeping track of Books we read, we would define a file type “Book” with common, well-known attributes like ISBN, author, title and keywords. This information is then either extracted by separate tools from ebook or PDF files, or manually entered by users, and stored in these well-defined attributes to be commonly accessible. A reference from a Note to a particular Book would then use the standard startOffset attribute as relation property, so SEN can use it for navigating to the specified position in the linked document. Books can now be searched and managed just like normal files, without the need for specialised library applications. Those are still useful for advanced use cases like import, export, or conversion, but not required for basic use cases. [13] Requiring users to configure a consistent set of file types, attributes and relations would counteract the goal of simplicity and ease of use.
因此，首选方法是使用定义明确、精心策划的本体配置，这些配置可以自由交换和导入。这使 SEN 能够依赖一组一致的实体和关系来进行 UI 和工具集成。
Therefore, the preferred way is to use well-defined, curated ontology configurations that can be freely exchanged and imported. This enables SEN to rely on a consistent set of entities and relations for UI and tool integration.
将 SEN 的概念模型映射到 Haiku 中的本机操作系统和桌面概念
Mapping the conceptual model of SEN to native OS and desktop concepts in Haiku
SEN 尽可能遵循文件系统语义，用户可以根据需要自由管理文件，也取决于用例和体验。文件夹是保存 PKG 的合理惯例，因为就像在普通桌面使用中一样，它们将相关信息组合在一起。这样，它们可以在用户之间轻松交换，或跨系统备份和恢复。
SEN adheres to file system semantics where possible, and users are free to manage their files as needed, also depending on the use case and experience. Folders serve as a sensible convention to hold a PKG, because like in normal desktop use, they group together related information. This way, they can be easily exchanged between users or backed up and restored across systems.
SEN 通过使用操作系统文件类型（见下文）来定义实体、用于保存其属性的属性以及具有名称、支持的关系源和目标（引用文件类型）等属性的可配置关系集，以及默认角色和标签，从而实现了一种简单的本体形式。[14]
SEN realises a simple form of ontologies by using operating system file types (see below) to define Entities, attributes for holding their Properties, and a configurable set of Relations with properties like name, supported relation sources and targets (referencing file types), and default roles and labels. [14]
然后，语义实体可以简单地表示为文件，并用一个有意义的图标表示现实世界（或虚拟）对象。这允许用户直接将他们对事物（如书籍、人物或地点）的心理模型映射到桌面对象，并以直接直观的方式与它们进行交互或操作。
Semantic Entities can then be simply represented as files, with a meaningful icon representing the real world (or virtual) object. This allows users to directly map their mental model of things (like books, people, or locations) to desktop objects and interact or manipulate them in a direct and intuitive manner.
实体不一定必须实际包含所表示对象的数据，它们可以充当占位符（或“代理”），可能（但不一定）根据需要指向实际对象。一个常见的例子是作为书签文件存储的 Web 链接，其中仅包含对在线网页的引用。在 Haiku 中，此类实体由具有自定义属性的空文件表示，其中包含引用或其他元数据（例如上次访问日期、网页源中定义的关键字或用户在创建书签时定义的关键字）。
Entities do not necessarily have to actually contain the represented object’s data, they can act as a placeholder (or “proxy”), possibly – but not necessarily – pointing to the actual object, as required. A common example is a web link stored as a bookmark file that contains only a reference to the online web page. In Haiku, such entities are represented by empty files with custom attributes containing references or other metadata (like date of last visit, keywords defined in the web page source or by the user when creating the bookmark).
在我们的示例中，我们可以将联系人实体表示为一个文件，其中包含 VCARD 格式的所有联系人数据，并将信息提取到文件属性中。这会产生一些冗余，可以通过仅保留存储在属性中的信息来解决。
In our example, we could have a contact entity represented as a file that contains all contact data in VCARD format, with the information extracted into file attributes. This creates some redundancy that could be resolved by keeping only the information stored in attributes.
这样，用户就可以从元级别构建知识图谱，描述事物及其之间的关系。实体与其内容分离，从而降低资源使用率并鼓励重复使用。在撰写内容时，无需在用户的桌面上（甚至在单独的 PKG 系统中）复制所有引用数据 - 例如，保存您的书籍或电影笔记。
This allows users to build knowledge graphs from a meta level, describing things and relations between them. Entities are decoupled from their content, which keeps resource usage low and encourages reuse. There is no need to replicate all referenced data on the user’s desktop (or even in a separate PKG system) when writing about it – e.g., keeping your book or movie notes.
更进一步，我们甚至可以将实际内容保存在远程位置，并通过其 ObjectID 引用文件，这些文件位于对象存储（例如 S3）中 [15]，或指向互联网资源 [16]（如语义网社区中使用的 IRI）、Solid pod（solidproject，2023）或使用 IPFS（IPFS，2023）。我们只需要将引用存储在单独的属性中，并提供一种解析目标的方法，就像在本地环境中一样——这仍在积极研究中，计划在项目后期进行。[17]
Going further, we could even keep the actual content in remote locations and reference files by their ObjectID in an object storage such as S3, [15] or pointing to internet resources [16] (like an IRI as used in the Semantic Web community), a Solid pod (solidproject, 2023), or using IPFS (IPFS, 2023). We just need to store a reference in a separate attribute and provide a way to resolve the target, just like in a local environment – this is still in active research and scheduled for a later project phase. [17]
除了上面提到的文件夹之外，通过 Context 提供了一种简单但更灵活的相关实体分组方法，Context 类似于标签，保存在由 SEN 管理的实体的文件属性 (SEN:CONTEXT) 中。这使用户可以摆脱桌面上传统的文件夹层次结构，以更合适的方式浏览他们的知识图谱。他们可以重复使用其他 PKG 并交叉引用类似想法的信息，而这些想法仍应分开保存。这可以防止 PKG 变得过于复杂，隐藏见解而不是提供见解——与 Nick Milo (Milo, 2022) 提出的“内容地图”概念相当，但同时更强大、更简单。无需手动创建包含上下文引用的“元注释”，并且由于 Context 属性已编入索引，用户可以将其包含在查询中，以发现给定主题、项目或领域或其组合的所有收集知识中的相关信息。举个例子，可以在 PKM 中搜索关于“哲学”、“政治”、“日本”和“江户”时代的任何内容。
In addition to folders mentioned above, a simple but more flexible way of grouping related entities is provided through a Context, which is like a tag and kept in a file attribute (SEN:CONTEXT) of an entity managed by SEN. This allows users to break free from traditional folder hierarchies on the desktop and navigate their knowledge graphs in a more suitable way. They can reuse other PKGs and cross-reference information for similar thoughts that should still be kept separately. This prevents the PKG from growing too complex, hiding insights instead of providing them – comparable to the concept of a “Map of Content” introduced by Nick Milo (Milo, 2022), but more powerful and easier at the same time. There is no need for manually creating a “meta-note” that contains contextual references, and since the Context attribute is indexed, users can include it in queries to discover relevant information across all collected knowledge for a given topic, project or domain, or a combination thereof. An example would be searching the PKM for anything on “philosophy”, “politics”, “Japan” and the “Edo” epoch.
基于这一概念，SEN 将提供“焦点视图”，将建议用于链接和导航的实体和关系限制在给定上下文中。这允许用户专注于特定的 PKG，但在需要时可以接触到全局实体和关系集。
Building on this concept, SEN will provide a “focus view”, limiting entities and relations proposed for linking and navigation to a given Context. This allows users to focus on a particular PKG but reaching out to the global set of entities and relations when needed.
Haiku 使用标准 MIME 类型来识别文件类型，如 Haiku FileTypes 用户指南 (Humdinger, 2009) 中所述。[18]
Haiku uses standard MIME types for identifying file types, as described in the Haiku FileTypes user guide (Humdinger, 2009). [18]
官方注册的 MIME 类型列表 (Melnikov, 2022) 涵盖了各种媒体类型（因此也涵盖了实体类）。为了在知识图谱中更通用地使用，我们需要添加自定义实体类型，可以使用 application/vnd.* 命名空间将其定义为有效的 MIME 类型。
The list of officially registered MIME types (Melnikov, 2022) covers a wide variety of media types (and hence entity classes). For a more generic use in a knowledge graph, we will need to add custom entity types, which can be defined as valid MIME types using the application/vnd.* namespace.
然后，文件属性保存实体属性，可以使用标准 FileTypes 设置应用程序完全根据用户的需求进行定制。
File attributes then hold Entity properties, which can be customised completely to the user’s needs by using the standard FileTypes settings application.
所有这些概念都被组合起来并在文件类型、属性、索引和查询（Humdinger，2019）的研讨会上投入使用，该研讨会描述了这里列出的概念在实际用例中的应用。
All of these concepts are combined and put to use in a workshop on file types, attributes, index and queries (Humdinger, 2019), which describes the application of the concepts laid out here to a real-world use case.
如前所述，只有关系及其属性不受 Haiku 或其文件系统（或当今任何其他操作系统）的原生支持，因此 SEN 保留单独的配置来管理关系并定义它们可以链接的实体（使用 MIME 类型），以及支持的关系属性和可能的别名，以便它们可以被重用和翻译。
As stated earlier, only Relations and their properties are not natively supported by Haiku or its filesystem (or any other OS today), so SEN keeps a separate configuration to manage relations and to define what Entities they can link (using MIME types), as well as supported relation properties and possible aliases, so they can be reused and translated.
有了这个技术基础和精心设计的扩展，我们最终可以将构建个人知识图谱所需的所有语义概念映射到用户的桌面上。
With this technical foundation and carefully designed extensions, we can finally map all the semantic concepts needed for building a personal knowledge graph onto the user’s desktop.
使用 SEN 实现知识图谱的可视化
Implementation and Visualisation of Knowledge Graphs with SEN
如上所述，SEN 将常见且成熟的桌面隐喻（如文件和文件属性）映射到语义概念，作为用户熟悉的知名桌面环境的扩展。只有在必要时，这些概念才会被扩展以支持现有桌面操作系统所不具备的功能（例如，显示关系目标）。
As detailed above, SEN maps common and well-established desktop metaphors like files and file properties to semantic concepts, acting as an Extension to the well-known desktop environment users are familiar with. Only when necessary, these concepts are stretched to support functionality that is not part of existing desktop operating systems (e.g., for showing relation targets).
这样，用户就可以几乎透明地采用和利用语义概念，并在其基础上构建个人知识图谱。下一节将详细介绍 SEN 如何支持用户进行个人知识管理。
This way, users can adopt and utilise semantic concepts almost transparently, and build a personal knowledge graph on top of them. The next sections will provide more detail on how SEN works to support users in personal knowledge management.
在文件管理器“跟踪器”中可视化和导航关系
Visualising and Navigating Relations in the File Manager “Tracker”
由于 SEN 非常以用户为中心，不应仅限于专家和知识工作者，因此标准文件浏览器 Tracker (Haiku, Inc., 2022) 进行了扩展（经过最少的修改），以便关系在上下文菜单中可见，用户可以像打开普通文件一样打开相关文件，但增加了一些功能来支持语义关系，如图 6.6 所示。
Because SEN is very user-centric and should not be limited to experts and knowledge workers, the standard file browser, Tracker (Haiku, Inc., 2022), is extended (with minimal modifications) so that relations are visible in the context menu, and users can open related files just as they would with normal files, but with some added functionality to support semantic relations, as shown in figure 6.6.
在这里，用户将具有相关属性的食谱和配料作为独立集合保存，并通过关系将它们连接起来。从食谱导航到所需配料只需一个上下文菜单，使用选定配料查找食谱也是如此。
Here, the user keeps recipes and ingredients with relevant properties as independent collections and connects them via relations. Navigating from recipe to required ingredients is just a context menu away, and the same applies to finding recipes using selected ingredients.
图 6.6 在 Tracker 中导航关系
Figure 6.6 Navigating relations in Tracker
SEN API 为所有常见用例（如创建、导航和查询关系）提供了各种过滤器和变异，并为 Tracker 等应用程序提供了所需的功能。它也可以由简单的命令行工具使用，因为所有功能都是通过基于消息的脚本公开的（请参阅下一节）。
The SEN API provides various filters and mutations for all common use cases, like creating, navigating and querying relations, and provides applications like Tracker with the needed functionality. It can be also used by simple command line tools, because all functionality is exposed through message-based scripting (see next section).
为了方便用户交互，关系目标可以按实体类型（如作者、书籍、电影、人员）、角色（如参加者、贡献者、出版商、审阅者）或标签（如参加者、贡献者、引用）分组。
For convenient user interaction, relation targets can be either grouped by Entity type (like Author, Book, Movie, Person), role (like Attendee, Contributor, Publisher, Reviewer) or label (like attendedBy, contributesTo, cites).
在 Haiku（与 BeOS 一样）中，桌面隐喻已扩展为建立可点击菜单，以便直接从上下文菜单浏览文件夹及其内容。
In Haiku (as in BeOS), the desktop metaphor has been extended to establish clickable menus for navigating folders and their contents right from a context menu.
相同的比喻用于导航关系：单击“打开相关...”菜单中的任何关系将在单独的跟踪器窗口中打开该关系的所有目标，以及每个目标的关系属性。
The same metaphor is used for navigating relations: clicking on any relation in the “Open Related…” menu will open all targets of that relation in a separate Tracker window, along with relation properties for each target.
图 6.7 展示了一个简化的示例，说明用户如何浏览研究论文或讲义的所有参考文献，这些参考文献存储为简单文件，具有专用的文件类型来表示网页、PDF 文档、书籍或视频演示等实体。
The figure 6.7 shows a simplified example of how users could navigate all references of a research paper or lecture note, stored as simple files with dedicated file types to represent entities like web pages, PDF documents, books, or video presentations.
图 6.7 在 Tracker 中显示关系目标
Figure 6.7 Showing relation targets in Tracker
为了在专用的 Tracker 窗口中显示关系目标，SEN 会创建一个临时文件夹，其中包含指向实际文件的链接。这些链接将关系属性保存在自定义属性中，因此可以使用详细信息视图在 Tracker 的文件窗口中像普通文件属性一样显示和修改它们，类似于其他文件浏览器中的列视图。
For displaying relation targets in a dedicated Tracker window, SEN creates a temporary folder with links to the actual files. These links hold relationship properties in custom attributes, so they can be displayed and modified just like normal file properties in Tracker’s file window using a detail view, similar to column views in other file browsers.
因此，通过将选定的目标文件拖到关系目标窗口中，可以将新关系添加到文件中。这将在两个文件之间添加新链接，并且可以像编辑其他属性一样编辑关系属性。[19] 在此窗口中删除文件将删除与已删除目标文件的关系。
Consequently, new relations can be added to a file by dragging selected target files into the relation targets window. This would add a new link between the two files, and relationship attributes can be edited like other properties. [19] Deleting a file in this window would remove the relation to the deleted target file.
这种导航可以沿着关系链进行扩展，在文件浏览器内构建一个强大但简单的图形式导航：通过在上面的视图中对目标调用相同的“打开相关”操作，用户可以导航到该文件的所有关系，依此类推。[20]
This navigation can be extended along the chain of relations to build a powerful but simple graph-like navigation inside the file browser: by invoking the same “Open related” action on a target in the view above, users can navigate to all relations of that file, and so on. [20]
最后，除了通过打开此文件夹中的文件来访问关系实体之外，用户还可以使用包含上下文信息的关系属性直观地导航到处理该关系的应用程序内的关系目标。这样就可以打开 PDF 查看器并跳转到给定页面，或者打开 Web 浏览器并突出显示以 WebAnnotation 形式提供的引用文本，如上文的 Notes 图示例中所示。
Finally, in addition to accessing entities of a relation by opening files in this folder, users can intuitively navigate to the relation target inside the application that handles it, using relation properties that hold context information. This allows opening a PDF viewer and jumping to the given page, or opening a web browser with highlighting the text referenced provided as a WebAnnotation, as suggested in the Notes graph example above.
语义信息发现
Semantic Information Discovery
SEN 不仅允许用户通过扩展文件浏览器来导航语义链接，还可以根据上下文（见上文）和关系过滤信息。通过选择多个文件并调用关系菜单，Tracker（使用 SEN API）会呈现所选实体之间共享的关系的过滤视图。
SEN not only allows users to navigate semantic links through extending the file browser, but also filtering information based on context (see above) and relations. By selecting multiple files and invoking the relations menu, Tracker (using the SEN API) presents a filtered view on relations that are shared between the selected entities.
想象一下，选择多个联系人，然后获取他们参加过的常见会议或合著的书籍列表。对于食谱集，其中的配料与食谱相关联，选择几种配料将显示基于它们的所有食谱。
Imagine selecting multiple contacts and getting a list of common meetings attended, or books they have co-authored. For a recipe collection, where ingredients are linked to recipes, selecting several ingredients would show all recipes based on them.
这些是简单但功能强大的使用场景，目前现有的知识管理解决方案无法实现，甚至最新的已经从笔记转变为实体的解决方案也无法实现。用户不必学习新的工具和技术来捕捉和链接他们的想法和日常信息，而是可以使用他们已经从与图形桌面交互中了解的相同概念和方法。
These are simple but powerful usage scenarios that cannot be currently realised with existing knowledge management solutions, not even the most recent ones that have already transcended from notes to entities. Users don’t have to learn new tools and techniques just to capture and link their thoughts and everyday information, but can use the same concepts and methods they already know from interacting with a graphical desktop.
脚本和应用程序集成
Scripting and Application Integration
这种无缝集成通过 Haiku 中的扩展脚本功能支持，它提供了基于消息的通信和扩展机制，用于控制系统本身的各个方面以及所有本机 Haiku 应用程序。
This seamless integration is supported through the extended scripting functionality in Haiku, which provides a message-based communication and extension mechanism for controlling various aspects of the system itself, as well as all native Haiku applications.
对于诸如GET、OPEN或CLOSE之类的简单操作，有明确定义的标准消息，但应用程序可以支持任何类型的自定义消息，然后可以将其捆绑在“脚本套件”的形式中。
There are well-defined standard Messages for simple operations like GET, OPEN or CLOSE, but applications can support any kind of custom messages, which can then be bundled in the form of “scripting suites”.
为了支持上述导航，需要扩展标准 PDF 查看器以支持在给定页面上打开文件（例如，通过向已支持的标准 OPEN 消息添加属性页面）。Web 浏览器需要支持 WebAnnotations 以及以类似方式为给定 URL 显示它们的方法，例如，使用附加属性（如 OPEN 消息的注释）。
To support the kind of navigation above, the standard PDF viewer needs to be extended to support opening a file on a given page (e.g., by adding a property page to the standard OPEN message already supported). The web browser needs to support WebAnnotations and a way to show them for a given URL in a similar way, e.g., using an additional attribute like annotation for the OPEN message.
SEN 用例示例
SEN Use Case Examples
最后，让我们看看 SEN 是如何实际使用的。以下示例说明了用户可以在日常桌面工作中应用知识图谱原理的可能使用场景。
Let’s finally see how SEN is used in action. The following examples illustrate possible usage scenarios where users can apply principles of knowledge graphs in their daily desktop work.
示例：书籍、作者和出版商的个人知识图谱
Example: A Personal Knowledge Graph of Books, Authors and Publishers
作为一名作家，我们会从三个实体构建一个图谱：书籍、作者和出版商。我们想要模拟不同的角色和联系：作者写书，由出版商审阅和出版，出版商最终向作者支付费用。
As a writer, we would build a graph from three entities: Book, Author and Publisher. We want to model different roles and connections: the Author writes a Book that is reviewed and published by a Publisher who eventually pays the Author.
有关关系的详细信息（如出版日期或付款金额）被建模为关系属性。请注意，此示例中的自定义关系属性是可选的，只有 SEN 属性（以 SEN: 开头）是必需的，因此我们可以通过 ID 连接实体。此外，关系属性仅对每个关系是唯一的，如关系 Publisher→Book 中所示，它们具有单独的角色。这样，我们可以对 Publisher 参与 Book 的各种方式进行建模。见图 6.8。
Details about the relations (like date of publication, or amount of payment) are modelled as relationship attributes. Note that the custom relationship attributes in this example are optional, only SEN attributes (starting with SEN:) are mandatory, so we can connect entities via IDs. Also, relationship properties are only unique per relation, as shown in the relations Publisher→Book, which have separate roles. This way, we can model the various ways in which the Publisher is involved in the Book. See figure 6.8.
图 6.8 图书图表
Figure 6.8 Book Graph
在上面的例子中，书籍章节存储在名为“Book of SEN.md”的文件中。
In this example illustrated above, a book chapter is stored in a file named “Book of SEN.md”.
该文件为文本文件类型，MIME 类型为 text/markdown，代表书籍实体。为简洁起见，此处仅显示（可选）SEN:TYPE。
The file is of type text file with MIME-type text/markdown, which represents a Book entity. For brevity, only the (optional) SEN:TYPE is shown here.
为了将一个人与一本书联系起来，SEN 注册了一个与标签作者和属性角色的关系。标签描述了关系，而角色则附加在关系源上。由于关系始终是双向的，因此可以建立两个连接实体的角色。
For connecting a Person to a Book, SEN has registered a relation with label authors and a property role. While the label describes the relation, the role is attached to the relation source. Since relations are always two-directional, the roles of both connected entities can be established.
文件“Gregor Rosenauer”的类型为 application/x-person，表示人员，仅包含强制性的 SEN 标准属性和文件名。此人是相关书籍的作者。
File “Gregor Rosenauer” of type application/x-person, which denotes a Person, only contains the mandatory SEN standard attributes and the file name. This person acts as an author of the related Book.
文件“Writer's Block”类型为 application/x-person，代表充当出版商、审查和出版书籍的组织。
File “Writer’s Block” of type application/x-person, represents an Organisation that acts as a Publisher, reviewing and publishing the Book.
示例：注释和注解
Example: Notes and Annotations
此示例显示了引用书籍的 Note 实体，其中注释引用了其中的章节（为简洁起见，省略了文件扩展名）。见图 6.9。
This example shows a Note entity referencing a Book with annotations referencing sections therein (file extensions omitted for brevity). See figure 6.9.
图 6.9 书籍笔记图
Figure 6.9 Book Notes Graph
文件“Plato – Symposion.pdf”的类型为 application/pdf（代表书籍实体），包含这部经典著作的全文。文件“Eros.md”的类型为 text/markdown，代表通过偏移量引用书中特定文本段落的注释实体。
File “Plato – Symposion.pdf” is of type application/pdf (representing a Book entity) and contains the full text of this classic work. File “Eros.md” of type text/markdown and represents a Note entity that references a particular passage of text in the book by offset.
文本范围以简单的基于索引的关系属性 startoffset 和 endOffset 的形式提供，但将更接近 WebAnnotation（W3C，2017）的建模，因此可以以一致的方式支持不同的媒体类型。
The text range is provided as simple index-based relationship attributes startoffset and endOffset, but will be modelled closer to WebAnnotation (W3C, 2017), so different media types can be supported in a consistent way.
例如：日历
Example: Calendar
我们可以通过引入事件文件类型来构建一个简单但集成良好且可见的日历，它可以包含标准 ICS 格式的事件数据（Desruisseaux，2009）。我们甚至可以定义基于顺序和时间的连接事件关系（在重复事件之间或每日/每周议程中的事件之间导航）。
We can build a simple but well integrated and visible calendar by introducing an Event file type, which could contain event data in standard ICS format (Desruisseaux, 2009). We can even define relations for connecting events based on sequence and time (navigating between recurring events or between events in the daily/weekly agenda).
然后，我们可以从查询给定日期的所有事件中获得一个简单的“今天”视图，按标签、参与联系人或位置或与关系相关的任何实体文件进行过滤。
We can then have a simple “Today” view from a query for all Events on a given date, filtered by tags, participating contacts, or Location, or any entity file connected with a relation.
单击桌面栏（Haiku 的应用程序启动器和信息面板）中的“日历”图标将打开一个跟踪器文件视图，其中包含所有事件日期为今天的事件文件，使用图 6.8 所示的关系目标视图。
Clicking on a “calendar” icon in the desk bar (Haiku’s application launcher and info panel) would open a Tracker file view with all event files having an event date of today, using the relation target view illustrated in figure 6.8.
可以使用从事件数据（对于 ICS 内容）或手动创建中收集的内在关系来检查单个事件，如图 6.10 所示。
Individual Events can be inspected using intrinsic relations gathered from the event data (for ICS content) or from manual creation, as in figure 6.10.
图 6.10 日历浏览器
Figure 6.10 Calendar Browser
笔记：UNO – 通用笔记本的概念
Notes: UNO – a Concept for a Universal NOtebook
这是一个高级用例的示例，其中应用程序使用 SEN API 提供语义链接和实体提取。
This is an example of an advanced use case, where an application is using the SEN API to offer semantic linking and entity extraction.
在之前奠定的基础上，利用 SEN，我们可以将一个简单的文本编辑器扩展为一个语义链接的笔记系统，提取元数据和对实体的引用作为文件系统中的关系。用户可以独立于编辑器发现、修改和浏览生成的知识图谱，但可以在需要时使用它，例如，为了获得图 6.11 中所示的集成用户体验。
Building on the foundations laid out before and utilising SEN, we can extend a simple text editor into a semantically linked note-taking system, with extracting metadata and references to Entities as relations in the file system. Users can discover, modify and navigate the resulting knowledge graph independently of the editor, but use it whenever they need to, e.g., for the integrated user experience depicted in figure 6.11.
图 6.11 具有 SEN 的语义笔记应用程序的原型概念
Figure 6.11 Prototype concept of a semantic note-taking application with SEN
这将解决常见笔记记录中的一个主要难题，即用户经常面临“收集者谬误”——积累永远不会再使用的信息（Giaro，2022 年）。随着 SEN 与文件系统和桌面紧密集成，笔记永远不会消失，并且通过为常见桌面工作提供自然扩展来促进重复使用。笔记将显示在搜索中，关系可直接显示在文件浏览器中。
This would solve one major dilemma in common notes taking, where users often face “collector’s fallacy” – building up information that is never used again (Giaro, 2022). With SEN integrating tightly with the file system and desktop, notes never get out of sight, and reuse is facilitated by providing natural extensions to common desktop work. Notes will show up in search, and relations are directly visible in the file browser.
这种不断重新审视和改进笔记的方法也可以与 (Matuschak, nd) 中引入的“常青笔记”进行比较。
This approach of constant revisiting and evolving notes can be also compared to “Evergreen Notes” introduced in (Matuschak, n.d.).
格式可以是 MarkDown，或者编辑器可以采用 Org 模式（Worg，2023）来实现更高级的语义功能，同时仍然是纯基于文本的。[21]
The format could be MarkDown, or the editor could adopt Org mode (Worg, 2023) for more advanced semantic features, while still being purely text based. [21]
此示例展示了 SEN 的强大而简单且集成良好的方法：笔记保存在标准文本文件中，但仍可以高效快速地查询和组织。相比之下，专有笔记系统通常以自定义格式锁定数据，并需要数据库作为存储后端。例如，捆绑的笔记应用程序 MacOS Notes 让用户难以轻松访问和备份他们的数据（Horowitz，2020 年）。
This example shows the powerfully simple and well-integrated approach of SEN: notes are kept in standard text files but can still be queried and organised efficiently and quickly. By contrast, proprietary note-taking systems often lock data in custom formats and require a database as storage backend. E.g., MacOS Notes, the bundled notes app, makes it hard for users to easily access and backup their data because of this (Horowitz, 2020).
如上所述，可以通过在文件浏览器中浏览文本文件的关系来独立浏览注释和所有提取的实体。见图 6.12。
As outlined above, notes and all extracted entities can be browsed independently by navigating the text file’s relations in the file browser. See figure 6.12.
图 6.12 使用 SEN 在文件浏览器中浏览语义注释
Figure 6.12 Navigating semantic notes in the file browser with SEN
该语义编辑器还适用于故事写作、映射和引用人物、地点和故事情节作为（自我）引用。
This semantic editor would also be suitable for story writing, mapping and referencing characters, locations and the story arch as (self-)references.
电影参考
Movie References
在这个高级示例中，我们将使用SEN进行影片分析，并展示三元关系的透明直观的导航。
In this advanced example, we will use SEN to do film analysis, and to show the transparent and intuitive navigation of ternary relations.
知名的在线服务（如 IMDb）提供了大量有关导演或其他电影的电影参考资料，但它们支持的实体有限（例如播放的歌曲或展示的艺术品），并且没有时间码来访问电影的参考部分。此外，所有参考资料都与 Web 界面紧密集成，并且没有简单的方法将它们导出到个人知识图谱中以供进一步使用。[22]
Well-known online services like IMDb provide a vast resource of movie references on directors or other movies, but they are limited in the entities they support (e.g., songs played or artwork shown), and there is no timecode to reach the referenced part of the movie. Also, all references are tightly integrated with the web interface, and there is no easy way to export them into a personal knowledge graph for further working with them. [22]
这里，一个简单的 Markdown 文档包含对演员、地点和歌曲的引用，以及通过研究电影获得的时间代码。
Here, a simple Markdown document holds references to actors, locations and songs, along with a time code obtained by studying the movie.
借助 SEN，用户可以浏览所有感兴趣的参考资料，发现相关实体及其出现的时间码。他们可以在文件浏览器中导航和查询它们，并在视频播放器中播放引用的片段，或者在地图中打开位置，如图 6.13 所示。
With SEN, users can then browse all references of interest, discover related entities and their timecode of appearance. They can navigate and query them in the file browser and play referenced segments in the video player, or open locations in a map, as illustrated in figure 6.13.
图 6.13 SEN 处理电影引用中的三元关系
Figure 6.13 SEN handling ternary relations in movie references
当前状态
Current State
目前，SEN 是一个个人研究项目，作为一个简单的概念验证，展示了 Haikus 文件浏览器 Tracker 中关系的使用。该项目将在基本愿景实现并在 GitHub 上提供后立即开源，同时提供大量文档和用例示例。一个简单的、不断发展的项目网站可在 (Rosenauer, 2023) 上找到，链接到目前仍为私有的源代码存储库。
In its current form, SEN is a personal research project acting as a simple proof-of-concept, showing the use of relations in Haikus’s file browser, Tracker. The project will be open sourced as soon as the base vision is implemented and made available on GitHub, along with extensive documentation and use case examples. A simple, evolving project web site is available at (Rosenauer, 2023), linking to the currently still private source code repository.
SEN 的最终目标是通过让高级用户直接在他们的桌面环境中处理他们的 PKG 来提供现有、独立的 PKG 工具的替代方案，而无需使用单独的定制工具以及固有的复杂性和成本。
The ultimate goal of SEN is to offer an alternative to existing, isolated PKG tools by enabling advanced users to work on their PKG directly in their desktop environment, without the need and inherent complexity and cost of using separate, custom tools.
结论
Conclusion
知识管理是一种固有用例，应该紧密集成到我们的日常工作环境中，而我们的日常工作环境仍然是桌面。当前的系统虽然很先进，但仍然引入了元级别，将信息与源分离，要求用户复制他们工作区中已有的文档和文件信息。
Knowledge management is an intrinsic use case that should be tightly integrated into our everyday working environment, which is still the desktop. Current systems, although advanced, still introduce a meta-level that decouples information from the source, requiring users to duplicate information they already have available on their workspace as documents and files.
通过在现有桌面操作系统中添加链接数据语义，所有收集和推断的知识都可以直接供用户使用。采用和扩展现有的桌面应用程序和隐喻使这些信息变得可操作，而不仅仅是可知。
By adding linked data semantics to an existing desktop OS, all collected and inferred knowledge becomes directly available to users. Adopting and extending existing desktop applications and metaphors makes this information actionable instead of just knowable.
由于现有桌面系统的限制，以及生态系统和社区方面的原因，采用 Haiku（一种小型且鲜为人知的操作系统）仍然很有意义。其基础、设计和技术使其成为实现 SEN 愿景的可行候选者。从那里开始，它可能会在将来适应其他系统，例如 Linux 上的 KDE，它在语义桌面领域已经拥有丰富的历史，但受到当前文件系统和工具支持的限制，因此不太适合用作
Because of constraints in existing desktop systems, and looking at the ecosystem and community aspect, it still makes much sense to adopt Haiku, a small and rather unknown operating system. The foundation, design and technology make it a viable candidate to realise the vision of SEN. From there, it may be adapted to other systems in the future, such as KDE on Linux, which already has a rich history in the field of semantic desktops, but suffers from limits in current filesystem and tool support that makes it less feasible to use as a foundation for
。
.
未来工作
Future Work
根据本章概述的发现，将构建一个更完整的原型，以实现上述更高级的用例。Tracker 将扩展以显示关系属性和目标，并使用 SEN 的 API 在引用位置打开文档或媒体文件。这将展示基于现有桌面操作系统中的语义桌面原则的个人知识图谱的工作概念，该操作系统提供现代文件系统功能（如查询）和自定义元数据支持。SEN 只需要提供用于在 Tracker 中操作和导航关系的扩展，包括跳转到引用的关系目标和使用支持的应用程序进行“深度链接”。然后，将使用该原型的反馈来推动进一步的开发。
Based on the findings outlined in this chapter, a more complete prototype will be built that enables more advanced use cases described above. Tracker will be extended to show relation properties and targets, and opening documents or media files at referenced positions by using SEN’s API. This will show a working concept of a personal knowledge graph based on semantic desktop principles in an existing desktop OS that provides modern file system features (like queries) and custom metadata support. SEN only needs to provide extensions for manipulating and navigating relations in Tracker, including jumping to referenced relation targets and “deep linking” using supported applications. Feedback from this prototype will then be used to drive further development.
动态关系
Dynamic Relations
在后期阶段，还可以在遍历时惰性地评估关系，例如在检索相关文件时。这对于昂贵且不稳定的关系（例如“类似”文件）非常有用，因为只有在知道关系源的情况下才能按需进行比较。例如，这将触发对基于指纹计算相似度的公式的评估。然后，用户可以直观地直接从文件浏览器中浏览类似的图片或歌曲。
In later stages, relations could also be lazily evaluated on traversal, e.g., when retrieving related files. This would be useful for expensive and volatile relations e.g., for “similar” files, where comparison can only be done on demand, when the relation source is known. This would trigger evaluation of a formula to calculate similarity based on fingerprints, for example. Users could then intuitively navigate similar pictures or songs directly from the file browser.
关系和属性的规则和推理
Rules and Inference of Relations and Attributes
与上述动态关系的惰性求值类似，推理可以通过使用基于文本的规则语言来实现，该语言存储具有关系配置的公理。规则引用具有关系源和目标或父文件夹占位符的属性。这样，SEN 可以评估传递关系并将结果写入属性，使结果在整个系统中透明可用，并根据上下文（文件夹）更改实体的属性。想象一下，用户联系人的人员关系通过家庭关系规则得到丰富，或者当具有相同的“公司”属性或位于同一个“公司”文件夹中时，工作联系人通过“同事”关系自动关联。
Similar to lazy evaluation of dynamic relations above, inference could be implemented, e.g., by using a text-based rules language storing axioms with the relation configuration. Rules reference attributes with placeholders for the relation source and target, or the parent folder. This way, SEN could evaluate transitive relations and write the result to attributes, making the result available transparently throughout the system, and change attributes of entities according to the context (folder). Imagine Person relations of user contacts being enriched by family relation rules, or work contacts being automatically related through a “coworker” relation when having the same “company” attribute, or living in the same “Company” folder.
更多语义跟踪器扩展
More Semantic Tracker Extensions
由于文件浏览器是 SEN 中个人知识图谱的主要界面，因此添加丰富的跟踪器视图可以大大增强语义桌面体验。例如，通过实现 2D 轴视图，实体可以排列在时间线上，或按接近度对齐等。
Since the file browser is the main interface to the personal knowledge graph in SEN, adding rich Tracker views could greatly enhance the semantic desktop experience. E.g., by implementing a 2D-axis view, Entities could be arranged on a timeline, or aligned by proximity etc.
在文件菜单中添加“操作”菜单将允许用户对实体进行简单的查看和导航以外的操作：这对于日历约会（“加入会议”）、电子邮件（回复）、联系人（呼叫）或位置（“导航方向……”）非常有用。SEN 会通过向其发送相应的脚本消息将命令移交给首选应用程序。
Adding an “Actions” menu to the file menu would allow users to operate on Entities beyond simple viewing and navigation: this could be useful for calendar appointments (“Join meeting”), E-Mails (Reply), Contacts (Call) or Locations (“Navigation directions…”). SEN would hand over the command to the preferred application by sending the respective scripting message(s) to it.
笔记
Notes
[1] 虽然不是一个新概念，但对文件系统属性的支持在不同的工具和操作系统中仍然有所不同。有些工具（如 UNIX 复制命令“cp”）需要特殊选项才能按预期工作，而有些工具（如“gedit”）则会默默删除它们。这就是 SEN 目前开发并专注于 Haiku 的原因之一，Haiku 是一种从头构建的现代操作系统，旨在为现代桌面提供元数据驱动的文件系统和桌面。
[1] Although not a new concept, support for filesystem attributes still varies among tools and operating systems. Some tools like the UNIX copy command “cp” require special options to work as expected, and some tools like “gedit” will silently delete them. This is one of the reasons why SEN is currently developed and focused on Haiku, a modern operating system built from scratch to provide a modern desktop with a metadata driven filesystem and desktop.
[2] Concept 网站上的介绍认为，文件不够灵活，限制太多，无法满足现代知识管理的需求，传统操作系统或文件系统和桌面系统都是如此。这正是 SEN 努力要改变的，但又不完全放弃桌面知识管理的优势。
[2] The introduction on the Concept web site argues that files are inflexible and too restricted to handle modern knowledge management needs, which is true for traditional operating or file systems and desktops. This is exactly what SEN strives to change, without abandoning the advantages of desktop-based knowledge management altogether.
[3] 与此最接近的功能之一是 MacOS 中的标签功能，它使用颜色标记文件（也可以重命名为更有用的分类，如“个人”、“工作”或“敏感”）。然而，这仅限于少数用例，并不旨在进行进一步的定制或扩展（甚至不支持添加标签）。
[3] One feature that comes closest to this is the usage of labels in MacOS to tag files with colours (which can also be renamed to more useful classifications like “Personal”, “Work” or “Sensitive”). However, this is restricted to few use cases and not intended for further customisation or extension (not even adding labels is supported).
[4] 例外的是，BSD 和 Darwin/MacOS 提供了扩展的 find 变体，支持分别使用 -xattr 和 xattrname 开关搜索扩展属性键和值。现在还有一个适用于 Linux 的开源工具 (Bhihe, 2022/2022)，它为带有 xattr 扩展的 find 命令提供了 shell 包装器。两者都只是命令行解决方案，不能充分满足我们的要求。
[4] As an exception, BSD and hence Darwin/MacOS provide an extended find variant that supports searching extended attribute keys and values using an -xattr and xattrname switch, respectively. There is now also an open source tool for Linux (Bhihe, 2022/2022) providing a shell wrapper for the find command with xattr extensions. Both are only command-line solutions and do not fulfil our requirements sufficiently.
[5] 基于完全以数据为中心的方法，该操作系统提供了基于插件的“数据转换器”，用于在外部数据格式和供本机应用程序使用的内部表示之间进行转换。
[5] Building on a fully data-centric approach, the OS provided plugin-based “data translators” for converting between external data formats and an internal representation for use in native applications.
[6] 请参阅https://www.haiku-os.org/docs/userguide/en/applications/people.html
[6] See https://www.haiku-os.org/docs/userguide/en/applications/people.html
[7] 请参阅https://www.haiku-os.org/docs/userguide/en/workshop-email.html
[7] See https://www.haiku-os.org/docs/userguide/en/workshop-email.html
[8] 现在回想起来，这或许是一个解决新平台“先有鸡还是先有蛋”难题的聪明方法，由于没有用户，新平台很难吸引开发者来开发高质量的软件，而用户又抱怨没有适合他们要求的像样的软件。
[8] In hindsight, it may have been a clever way to solve the chicken and egg dilemma of new platforms, where it is hard to attract developers to build quality software, since there are no users, and users complain about the lack of decent software to suit their requirements.
[9] 请参阅用户问题报告，例如 https://askubuntu.com/questions/1154499/baloo-creates-64gb-index-takes-half-my-memory-and-25-cpu
[9] See user problem reports such as https://askubuntu.com/questions/1154499/baloo-creates-64gb-index-takes-half-my-memory-and-25-cpu
[10] D-BUS 和 Qt 具有类似的基础架构，在 KDE（KDE eV，2023）中使用最为广泛，但 BeOS 和 Haiku 是围绕这一概念设计的，并提供了紧密集成的本机 API，可在整个系统中以相同的方式使用，包括在应用程序内。
[10] There is a similar infrastructure available with D-BUS and Qt used most prominently in KDE (KDE e.V., 2023), but BeOS and Haiku are designed around this concept and provide a tightly integrated, native API that is used throughout the system in the same way, also within applications.
[11] 基于消息的脚本是 BeOS 和 Haiku 中的标准概念，它提供了一个名为“hey”的命令行工具，用于通过消息传递与应用程序交互（类似于 macOS 中的“tell”命令）。
[11] Message-based scripting is a standard concept in BeOS and Haiku, which provides a command line tool called “hey” for interacting with applications using message passing (similar to the “tell” command in macOS).
[12] 现实地说，这在所有情况下都是不可能的，所以 SEN 将支持某种形式的别名，用于语义上相同的属性（例如，日期、金额、同一实体的不同拼写甚至本地化形式）。
[12] To be realistic, this will not be possible in all cases, so SEN will support some form of aliases for semantically equal attributes (e.g., dates, amounts, different spellings for the same entity or even localised forms).
[13] 这需要完全以数据为中心的系统，其中应用程序将数据用作可公开访问的资源。在这个例子中，我们可以构建一个混合解决方案，其中 Calibre 用于转换。由于提取的元数据存储在单独的数据库中，因此如果没有某种桥接，它就无法在应用程序之外重复使用。
[13] This requires fully data-centric systems, where applications use data as openly accessible resources. In this example, we could build a hybrid solution where Calibre is used for conversion. Since extracted metadata is stored in a separate database, it could not be reused outside the application without some bridging.
[14] 之后，可以从 RDF 或 JSON-LD 形式的现有定义中导入这样的本体，其中通用语义标准中定义的类及其属性根据文件类型和属性映射到 SEN 的本机表示。
[14] Later, such an ontology could be imported from an existing definition in RDF or JSON-LD form, where classes and their properties defined in a universal semantic standard are mapped to SEN’s native representation based on file types and attributes.
[15] 遗憾的是，当前的远程文件系统（例如 Samba 或 NFS）不允许访问元数据，而元数据是以与操作系统和位置无关的方式寻址文件所必需的。
[15] Sadly, current remote file systems such as Samba or NFS do not allow accessing metadata, which is required to address files in an OS and location agnostic way.
[16] 另请参阅（Brander，2022）中“智慧圈”的概念和原型。
[16] See also the concept and prototype of a “noosphere” in (Brander, 2022).
[17] 我们可以包含对 S3 ObjectID、IRI 或 IPFS 情况下的 CID 的引用。
[17] We can include a reference to the S3 ObjectID, an IRI, or a CID in the case of IPFS.
[18] 在其他操作系统中使用的文件扩展名在 Haiku 中完全是可选的，因为操作系统使用 MIME 数据库并检查文件的类型属性是否匹配。文件类型是通过检查文件本身来检测的，使用 MIME 类型嗅探，例如解析 magic bye。扩展名只是在其他所有方法都失败的情况下的后备方法。
[18] File name extensions as used in other operating systems are purely optional in Haiku, as the OS uses a MIME database and checks the file’s type attribute for a match. The file type is detected by inspecting the file itself, using MIME type sniffing, e.g., parsing magic byes. Extensions are only a fallback, if all else fails.
[19] 这些更改（全部发生在临时视图中）会传播到关系中涉及的实际文件，并从中传播。SEN 将关系属性以键/值形式存储在单个文件属性中。
[19] These changes (all happening in a temporary view) are propagated to and from the actual files involved in the relation. SEN stores relationship attributes in a single file attribute in key/value form.
[20] 扩展后的跟踪器将提供一种“SEN 模式”，其中仅公开语义功能，因此用户可以更轻松地导航关系，就像在图表中一样，例如，通过在文件上下文菜单中仅显示关系。
[20] The extended Tracker will provide a “SEN mode”, where only semantic functionality is exposed, so users can navigate relations more easily, like in a graph, e.g., by showing only relations in the file context menu.
[21] 最初作为 Emacs 的编辑器模式提供，它已被改编为各种用例并与许多环境集成，甚至可以链接到应用程序和数据，例如参见 https://orgmode.org/worg/org-mac.html
[21] Originally provided as an editor mode for Emacs, it has been adapted for various use cases and integrated with many environments, even linking to applications and data is possible, see for example https://orgmode.org/worg/org-mac.html
[22] 最著名的电影数据库 IMDb 需要一个 AWS 账户，网络流量按使用量计费。Letterboxd 是一个免费的替代方案，它提供了一个开放的、免费使用的 API，但功能也有限。
[22] The most prominent movie database, IMDb, requires an AWS account where network traffic is billed by usage. Letterboxd is a free alternative that provides an open, free to use API, but also has limited functionality.
用例、原型和实现
Use cases, prototypes, and implementations
第七章
Chapter 7
图表不是关键，而是让我们找到关键的东西
Graphs aren’t the thing, they’re the thing that gets us to the thing
马蒂纳斯·尤塞维丘斯
所谓的个人知识图谱工具在过去几年中如雨后春笋般涌现。尽管流行语将它们比作“网络思维”和“第二大脑”，但它们通常只不过是碰巧使用某种形式的图形数据模型的笔记软件。
So-called Personal Knowledge Graph tools have mushroomed over the last few years. Despite catchphrases likening them to “networked thoughts” and a “second brain”, they are usually not much more than note-taking software that happens to use some form of the graph data model.
这些工具最突出的特点是可寻址的内容块和双向链接。内容块具有隐式标识符，可以从其他内容块引用，从而在它们之间创建有向但未标记的关系。然后，该工具推断出逆向关系。因此，可以将这些关系可视化为有向无标记图。数据通常存储在 Markdown 文件中。
The most prominent features of these tools are addressable content blocks and bidirectional links. Content blocks have implicit identifiers that can be referenced from other content blocks, thus creating a directed, but unlabeled, relationship between them. The tool then infers the inverse relationship. The relationships can thus be visualized as a directed unlabeled graph. The data is usually stored in Markdown files.
我认为这只是个人知识图谱概念的一个非常有限的版本。我将从广义上回顾个性化图谱的一些历史用法，然后提供广义定义并确定其关键特征。此外，我将介绍广义 PKG 的实用概念验证实现，以解决当前的局限性。
I believe that is a very limited version of the Personal Knowledge Graph concept. I will look at some historical usages of personalized graphs in a broad sense before providing a generalized definition and identifying their key traits. More than that, I will present a working proof-of-concept implementation of generalized PKGs that addresses the current limitations.
对于不熟悉语义网标准或需要复习的读者，本章末尾提供了一个附件，其中包含有关资源描述框架 (RDF)、词汇表和本体以及存储和查询 RDF 图的信息。此附件由 Omes Baltes 和 Maribel Acosta 提供。
As a reference for readers not familiar with Semantic Web standards, or in need of a refresher, an annex is provided at the end of this chapter with information on the Resource Description Framework (RDF), vocabularies and ontologies, as well as storing and querying RDF graphs. This annex has been contributed by Omes Baltes and Maribel Acosta.
软件代理
Software agents
2001 年的著名文章《语义网》（Berners-Lee 等，2001）谈到了一种软件“代理”，它可以根据指定的标准（例如“距离家 20 英里半径范围”和“优秀或非常好的评级”）自动预约医生。该代理能够从医生办公室的另一个代理那里检索医疗记录。
The famous 2001 article “The Semantic Web” (Berners-Lee et al., 2001) talks about a software “agent” which automatically books doctor appointments according to specified criteria such as “20-mile radius from home” and “excellent or very good ratings.” The agent is able to retrieve medical records from another agent at the doctor’s office.
尽管这篇文章是在“知识图谱”这一术语诞生之前发表的，但我们可以说这篇文章描述了个人知识图谱用例。本文还提供了语义网技术堆栈的高级概述，其中 URI 标识符和 RDF 数据模型是基础组件。20 年后，经过多次更名，它仍然是支持语义知识图谱的完全相同的技术堆栈。
Even though it was published well before the term “Knowledge Graph” was conceived, we can argue that the article describes a personal Knowledge Graph use case. The article also offers a high-level overview of the Semantic Web technology stack, with URI identifiers and the RDF data model as the foundational components. 20 years later and having gone through multiple rebrandings, it’s still the exact same technology stack that powers semantic Knowledge Graphs.
朋友的朋友
Friend of a Friend
FOAF (Brickley & Miller, 2004) 是早期的 RDF 词汇表，用于描述人员、人员活动以及人员与其他人和对象之间的关系。它是当今使用最广泛的词汇表之一。
FOAF (Brickley & Miller, 2004) is an early RDF vocabulary for describing persons, their activities and relations to other people and objects. It is one of the most widely used vocabularies today.
FOAF 可用于实现社交网络的去中心化、机器可读的图形表示。人员实例使用基本的“知道”关系连接。网页和主页、Messenger 帐户、图像和项目等概念也可以描述并附加到图形中。
FOAF can be used to implement decentralized, machine-readable graph representations of social networks. Instances of persons are connected using the basic “knows” relation. Concepts such as webpages and homepages, messenger accounts, images, and projects can be described and attached to the graph as well.
个人资料储物柜
Personal data locker
图 7.1 个人数据储物柜愿景 (Siegel, 2016)
Figure 7.1 Personal Data Locker Vision (Siegel, 2016)
David Siegel 提出了个人数据保险箱的设想（Siegel，2009）。数据包括日历事件、约会、家庭成员、个人新闻、朋友和同事的更新、按优先级排序的任务列表，甚至实验室测试结果、金融投资、汽车数字证书等。它分为五类：家庭、库存、金融、媒体和医疗。
David Siegel proposes a vision of a personal data locker (Siegel, 2009). The data includes calendar events, appointments, family members, personal news, updates from friends and colleagues, a task list sorted by priority, even lab test results, financial investments, digital certificate for the car, etc. It is grouped into five categories: home, inventory, financial, media, and medical.
数据柜可以连接到外部来源，例如连接到医生办公室以便预约。
The data locker can connect to external sources, for example to the doctor’s office in order to book appointments.
谷歌的研究议程
Google’s research agenda
图 7.2 个人知识图谱图示（Balog & Kenter，2019）
Figure 7.2 Illustration of a personal knowledge graph (Balog & Kenter, 2019)
Google 的研究议程 (Balog & Kenter, 2019) 将 PKG 定义为与其用户（即中心实体）个人相关的实体图。这些实体可以包括相关人员（“妈妈”）、拥有的物品（“我的吉他”）、位置（“家乡”）。它可能支持语义分类（使用分类法）以及推理（使用本体）。它还设想链接到外部数据源中的实体（链接数据）。
Google’s research agenda (Balog & Kenter, 2019) defines PKGs as a graph of entities personally related to its user, which is the central entity. Those can include related people (“mom”), owned things (“my guitar”), locations (“hometown”). It may support semantic categorization (using taxonomies) as well as inference (using ontologies). It also envisions linking to entities in external data sources (Linked Data).
这篇论文没有提供 PKG 模型的实现，但确实包含了多个与 RDF 相关的参考资料。它还讨论了潜在的用途，即智能个人助理和个性化聊天机器人。在这方面，它在概念上与 2001 年的“语义网”文章相关，其中指出：
The paper does not offer an implementation of the PKG model, but it does include multiple RDF-related references. It also discusses potential utilization, namely intelligent personal assistants and personalized chatbots. In that regard it is conceptually related to the 2001 “The Semantic Web” article, stating:
事实上，很难想象没有 PKG 的真正个人助理。
In fact, it is difficult to imagine a truly personal assistant without a PKG.
广义 PKG
Generalized PKGs
我们看到，个人知识图谱的广泛用途涉及虚拟和现实世界实体，通过不同类型的关系将人作为中心实体连接起来。个人知识图谱作为相互关联的个人笔记的当前含义要狭窄得多。
As we can see, the broad usage of personal Knowledge Graphs involves both virtual and real-world entities, connected by different types of relations to the person as the central entity. The current meaning of PKGs as interconnected personal notes is much narrower.
我们如何才能协调 PKG 的广义和狭义用法？让我们先了解一下基本原理，然后概括定义以适应两者。
What can we do to align the broad and the narrow usage of PKGs? Let’s look at first principles and generalize the definition to accommodate them both.
我们研究的用例包含各种实体类型，有些是重叠的（事件、事物），有些则更为专业（投资、票据）。这引出了广义 PKG 系统的第一个特征：在实体类型方面，它应该是开放式的（通用的）和面向未来的，即，它不应该被硬编码为仅支持一组有限的实体类型，而应该允许用户根据需要添加新类型。
The use cases we looked at contain a variety of entity types, some overlapping (events, things) and some more specialized (investments, notes). This leads us to the first trait of a generalized PKG system: it should be open-ended (generic) and future-proof when it comes to entity types, i.e., it should not be hardcoded to only support a limited set of entity types but allow the user to add new types on demand.
这种系统的第二个特点是，它需要图形模型中的带标签关系，而不是基于 Markdown 的工具所使用的无标签关系。如上文中“地址”或“知道”等简单示例所示，带标签的关系是一等公民，具有自己的语义，这使得图形数据模型更加丰富。这种数据模型支持跨应用程序共享含义，非常适合知识表示和推理，其中语义规则可以应用于现有知识以推断新知识。
The second trait of such a system is that it would require labeled relations in the graph model, not the unlabeled ones that the Markdown-based tools are using. Illustrated by simple examples I have shown above such as “address” or “knows”, labeled relations are first-class citizens and carry semantics of their own, which allow for a much richer graph data model. Such a data model enables shared meaning across applications and is well suited for knowledge representation and reasoning, where semantic rules can be applied to existing knowledge in order to infer new knowledge.
如果有人想将标记的关系作为未标记的关系呈现给用户，这很容易做到——但反之则不行。
And if one wants to present a labeled relation to the user as an unlabeled one, it can easily be done – but not the other way around.
第三个特征与限定词“个人”有关。从数据的角度来看，尽管用户在概念上处于中心位置，但他们只是图谱中的另一个实体。因此，我们可以说，个人知识图谱与其他知识图谱一样，只是它以用户为中心呈现给用户。
The third trait pertains to the qualifier “personal.” From the data perspective, despite conceptually being at the center of it, the user is just another entity in the graph. We can therefore argue that a personal Knowledge Graph is just a Knowledge Graph like any other, but it is presented to the user as being centered around them.
如果人只是实体之一，笔记只是实体类型之一，那么是否有可能拥有某种通用的知识图谱框架或平台，其可定制性足以实现 PKG 用例？我坚信可以。但我们需要确保这样的系统足够灵活，能够支持上述特征以及笔记 PKG 工具的功能：可寻址块和它们之间的双向链接。
If the person is just one of the entities and a note is just one of the entity types, wouldn’t it be possible to have some kind of generic Knowledge Graph framework or platform that is customizable enough to implement the PKG use cases? I strongly believe so. But we need to make sure that such a system will be flexible enough to support the aforementioned traits as well as the features of the note-taking PKG tools: addressable blocks and bidirectional links between them.
RDF 实现
RDF implementation
看看上述用例的超集，很明显它的实现需要一个有向标记图。未标记的关系和 Markdown 文件格式不会满足要求。
Looking at the superset of the use cases above, it is clear its implementation would require a directed labeled graph. Unlabeled relations and the Markdown file format will not cut it.
为了实现最大程度的互操作性和面向未来的解决方案，我们应该采用标准技术。只有一个标准化的有向标记图数据模型：RDF，即资源描述框架。它拥有一整套建立在其之上的 W3C 规范（从序列化格式到本体、查询和约束语言），以及一个已经开发了 25 年的软件和数据集生态系统。请查看 W3C 语义网活动，了解有关链接数据、词汇表、查询、推理等的更多信息。
For maximum interoperability and future-proof solutions we should employ standard technologies. There is only one standardized directed labeled graph data model: RDF, the Resource Description Framework. It has a whole stack of W3C specifications built on top of it (from serialization formats to ontology, query, and constraint languages) as well as an ecosystem of software and datasets that has been developed for 25 years. Take a look at the W3C Semantic Web activity for more information on Linked Data, vocabularies, query, inference, etc.
RDF 堆栈不提供开箱即用的可寻址块和双向链接，我们需要自己建模和实现这些功能。
The RDF stack does not provide addressable blocks and bidirectional links out of the box, we need to model and implement these features ourselves.
可寻址块
Addressable blocks
图 7.3 片段标识符 (Connolly, 2003)
Figure 7.3 Fragment Identifiers (Connolly, 2003)
首先，什么是“块”？它是一个子文档级 UI 组件，具有 URI 标识符，就像 RDF 中的任何其他资源一样。更具体地说，块必须具有哈希 URI，即带有尾随片段标识符的 URI，它在 HTML 上下文中引用文档中的某个位置。另一个 HTML 文档中的链接可以引用该确切位置，它也可以用作 RDF 三元组中的主语或宾语。
First, what is a “block”? It is a sub-document level UI component that has a URI identifier, just like any other resource in RDF. More specifically, a block has to have a hash URI, i.e., a URI with a trailing fragment identifier, which in a HTML context refers to a location within a document. A link in another HTML document can reference that exact location, it can also be used as a subject or object in an RDF triple.
类型和关系
Types and relationships
要将音符块建模为 RDF，我们需要选择或创建一个本体，该本体将定义音符块的类型以及将它们相互关联的属性，或者可能混合和匹配来自多个本体的术语。这并不是试图规定一个最终取决于建模者的世界观的规范 RDF 模型，而是想出一个足以满足我们练习需要的东西。
To model note blocks as RDF, we need to choose or create an ontology that will define the types of the blocks and the properties that relate them to each other, or possibly mix and match terms from multiple ontologies. This is not an attempt to prescribe a canonical RDF model which ultimately depends on the worldview of the modeler but rather to come up with something that is good enough for the sake of our exercise.
我们可以轻松地将笔记块视为“思想单位”，这是 SKOS 本体中“概念”的定义，并使用 skos:Concept 类作为其 RDF 类型。至于将一个块与另一个块关联起来，我们可以重新利用 skos:related 属性 - 它是通用的和非特定的，类似于笔记工具中未标记的关系。skos:related 对我们的用例特别方便，因为它是对称的：<A> skos:related <B> 包含 <B> skos:related <A>。要描述块的内容，我们可以使用 SIOC 本体及其 sioc:content 属性。或者，如果我们决定将笔记视为信息资源（文档）而不是抽象概念，我们可以将 SIOC 用于整个模型而不是 SKOS。
We can easily treat note blocks as “units of thought,” which is the definition of a “concept” in the SKOS ontology, and use the skos:Concept class as their RDF type. As for relating one block to another, we can repurpose the skos:related property – it is general and unspecific, which is similar to the unlabeled relations in note-taking tools. skos:related is particularly convenient for our use case as it is symmetric: <A> skos:related <B> entails <B> skos:related <A>. To describe the content of a block we can use the SIOC ontology and its sioc:content property. Alternatively, we could use SIOC for our whole model instead of SKOS if we decided to treat notes as information resources (documents) rather than abstract concepts.
我们可以将块推广到文本之外，并假设它们可以表示任何类型的对象，例如音频和视频、表格、地图、 3D 模型、（可执行）源或查询代码等 - 任何可以在浏览器中呈现为块级元素的东西。
We can generalize blocks beyond text and presume that they can represent any type of objects, for example audios and videos, forms, maps, 3D models, (executable) source or query code, etc. – anything that can be rendered as a block-level element in the browser really.
图 7.4 可在项目中使用的交互式数据驱动块（Block Protocol，2022）
Figure 7.4 Interactive, data-driven blocks to use in your projects (Block Protocol, 2022)
丰富的、知识图谱支持的、人机可读的文档可以以完全数据驱动的方式由这些块组成。这是最近的块协议的主张，不幸的是，它不是基于 RDF 技术。这是一个有趣的方向，涉及低代码开发和结构化内容管理，但超出了本章的范围。
Rich, Knowledge Graph-backed, both human- and machine-readable documents could be composed from such blocks in a completely data-driven fashion. That is the proposition of the recent Block Protocol, which unfortunately is not based on RDF technology. It is an interesting direction that crosses into low-code development and structured content management but is beyond the scope of this chapter.
订购
Ordering
块按特定顺序出现在文档中。RDF 语句（三元组）是无序的，但 RDFS 规范确实提供了两种可用于排序的构造：序列容器和集合。
Blocks appear on a document in a certain order. RDF statements (triples) are unordered, but the RDFS specification does provide two types of constructs that can be used for ordering: sequence containers and collections.
序列使用名称中带有整数的属性来指示项目在序列中的位置：rdf：_1，rdf：_2，rdf：_3等。例如，序列中的第一个项目将被描述为_：seq rdf：_1 _：item。
Sequences use properties with integers in their name that indicate the position of the item in the sequence: rdf:_1, rdf:_2, rdf:_3 etc. For example, the first item in the sequence would be described as _:seq rdf:_1 _:item.
另一方面，集合需要更复杂的链表结构并使用 rdf:first 和 rdf:rest 属性，并以特殊的 rdf:nil 值终止列表。
Collections on the other hand require a more complex linked list structure and use the rdf:first and rdf:rest properties, with a special rdf:nil value terminating the list.
RDF 序列和列表之间的一个区别是，序列始终对新元素开放，而列表可以关闭。关闭块列表似乎对我们的用例没有用。
One of the differences between RDF sequences and lists is that sequences are always open for new elements while lists can be closed. Closing the list of blocks does not seem to be useful for our use case.
与列表相比，在序列中间插入元素可能需要更新更多语句，但更新操作更简单且性能更好（Daga，2019）。
Inserting an element in the middle of a sequence can require updating many more statements compared to lists, but the update operation is much more simple and performs better (Daga, 2019).
与序列不同，列表可以组合，即一个文档中列表的中间部分可以链接到另一个文档中另一个列表的某个部分。虽然这可能用于实现包含，但必须通过遍历分散的片段来解析文档的结构，这代价高昂，而且在笔记 PKG 工具或一般的 Web 应用程序中不会出现这种情况。
Unlike sequences, lists can be composed, i.e., the middle of a list in one document can link to a section of another list in a different document. While this could potentially be used to implement transclusion, having to resolve the structure of a document by traversing decentralized fragments would be expensive, and is not something that is found in the note-taking PKG tools or web applications in general.
这些参数指向单个文档中包含的 RDF 序列，作为对有序块列表进行建模的更简单、更实用的方法。
These arguments point to RDF sequences contained within a single document as the simpler and more pragmatic way to model ordered lists of blocks.
例子
Example
根据我们的研究结果，我们最终可以编写一个完整的 RDF 示例（使用 Turtle 语法），该示例描述一个具有 2 个有序内容块的文档，其中第二个内容块链接到另一个文档中的块。
With our findings, we can finally compose a full RDF example (in Turtle syntax) that describes a document with 2 ordered blocks of content, the second one of them linking to a block in a different document.
@prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#> 。
@prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#> .
@前缀 skos： <http://www.w3.org/2004/02/skos/core#> 。
@prefix skos: <http://www.w3.org/2004/02/skos/core#> .
@prefix sioc: <http://rdfs.org/sioc/ns#> 。
@prefix sioc: <http://rdfs.org/sioc/ns#> .
<页面1#块1> rdf:_1 <页面1#块1>;
<page1#block1> rdf:_1 <page1#block1>;
rdf:_2 <page1#block2> 。
rdf:_2 <page1#block2> .
<page1#block1> 一个 skos:概念；
<page1#block1> a skos:Concept ;
sioc:content“这是块#1”。
sioc:content "This is block #1".
<page1#block2> 一个 skos:概念；
<page1#block2> a skos:Concept ;
sioc:content "这是块#2";
sioc:content "This is block #2";
skos：相关<page2#block1>。
skos:related <page2#block1> .
当然，我们可以为每个块附加更多的元数据，例如作者、创建或更新日期等等，但为了清晰起见，我将其保持简单。
Naturally we can attach much more metadata to each block, such as the author, date of creation or update and so on, but I’m keeping it simple for clarity.
双向链接
Bidirectional links
RDF 没有双向链接的概念：RDF 三元组中的属性是单向关系。这时，我们可以利用 RDF 查询语言 SPARQL，它是 RDF 技术栈的一部分。
RDF does not have a concept of bidirectional links: the property in an RDF triple is a one-way relationship. This is where we can make use of SPARQL, the RDF query language, which is part of the RDF technology stack.
CONSTRUCT 是 SPARQL 中的 4 种查询形式之一，可用于转换和扩充 RDF 数据集。它由两部分组成：对输入三元组进行模式匹配的 WHERE 子句，以及使用匹配结果中的变量绑定（如果有）创建新三元组的 CONSTRUCT 模板。
CONSTRUCT is one of the 4 query forms in SPARQL, it can be used for transforming and augmenting RDF datasets. It consists of two parts: the WHERE clause that pattern-matches the input triples, and the CONSTRUCT template that creates new triples using the variable bindings from the matching result (if there was any).
在支持 SPARQL 的实现中，我们可以使用以下 CONSTRUCT 查询来推断对称的 skos:related 关系——链接到其他块的块也会在相反方向上链接：
In a SPARQL-enabled implementation, we can use the following CONSTRUCT query to infer the symmetric skos:related relationship – blocks that link to other blocks also become linked in the opposite direction:
前缀 skos：<http://www.w3.org/2004/02/skos/core#>
PREFIX skos: <http://www.w3.org/2004/02/skos/core#>
构造
CONSTRUCT
{
{
？块？属性？对象。
?block ?property ?object .
?块 skos：相关？相关。
?block skos:related ?related .
?相关 skos：相关 ?块。
?related skos:related ?block .
}
}
在哪里
WHERE
{
{
？块？属性？对象。
?block ?property ?object .
选修的 {
OPTIONAL {
?块 skos：相关？相关
?block skos:related ?related
}
}
}
}
这是一个简单的例子，说明如何使用 SPARQL 代替推理器从基础数据推断出新的三元组。在笔记工具的 RDF 实现中使用这种方法，用户只需创建单向关系，但系统会在检索时动态将其呈现为双向关系。查询性能主要取决于数据集的大小和特定的 RDF 数据库。或者，可以使用相同的查询预先计算（具体化）反向链接，这将提高检索性能，但会使更新机制更加复杂。
This is a simple example of how SPARQL can be used instead of reasoners to infer new triples from ground data. Using this approach in the RDF implementation of a note-taking tool, the user would only have to create a one-directional relationship, but the system would present it as a bidirectional one on the fly upon retrieval. The query performance would depend mostly on the volume of the dataset and the specific RDF database. Alternatively, the backlinks could be precomputed (materialized) using the same query, which would improve the retrieval performance but make the update mechanism more complicated.
链接数据中心
LinkedDataHub
我曾声称，通过定制一个支持可寻址块和双向链接的通用知识图谱平台，应该可以实现广义的 PKG 用例。作为证明，我们已经实现了这样一个平台。
I have claimed that it should be possible to implement the generalized PKG use case by customizing a generic Knowledge Graph platform that supports addressable blocks and bidirectional links. As a proof, we have implemented such a platform.
LinkedDataHub 的用户体验并不以用户为中心，但基于声明性技术的前端框架适合低代码开发，使得默认 UI 的大量定制和扩展也比从头开始构建 UI 更具成本效益。
LinkedDataHub’s UX is not centered around the user’s person, but the frontend framework based on declarative technologies lends itself to low-code development, making even substantial customizations and extensions of the default UI significantly more cost effective than having the UI built from scratch.
LinkedDataHub 支持基于块的内容范例以及双向链接（称为“反向链接”）。它们的实现方式与本文所述完全相同，并且适用于所有类型的实体。
LinkedDataHub supports both the block-based content paradigm as well as the bidirectional links, which are called “backlinks”. They are implemented exactly as described in this article and work with all types of entities.
图 7.5
Figure 7.5
图 7.6
Figure 7.6
LinkedDataHub 是一个中间件，可以在兼容 SPARQL 1.1 的三重存储上运行，它使用三重存储来保存 RDF 数据。LinkedDataHub 中的每个文档也是图形存储中的一个命名图形，可以使用任何标准 RDF 语法以及 HTML 表示来提供服务。该模型是本体驱动的：更改用户可以创建和编辑实例的类以及这些类的属性和关系不需要任何编程，可以完全使用 UI 或 CLI。UI 使用专门的声明性样式表进行呈现。
LinkedDataHub is middleware that can run on top of a SPARQL 1.1-compatible triplestore, which it uses to persist RDF data in. Every document in LinkedDataHub is also a named graph in the graph store and can be served using any standard RDF syntax as well as the HTML representation. The model is ontology-driven: changing the classes users can create and edit instances of, as well as the properties and relationships of those classes, does not require any programming and can be entirely using the UI or the CLI. The UI is rendered using exclusively declarative stylesheets.
我相信，借助 LinkedDataHub，我们已经找到了通用平台的数据驱动、基于标准和 RDF 原生的软件架构，该架构可用于实现 PKG 软件以及越来越多地使用语义 RDF 知识图谱的各个行业中任意数量的特定领域用例。与从头开始构建应用程序相比，扩展和自定义 LinkedDataHub 的域和 UI 组件来实现此类用例要高效得多。
I believe that with LinkedDataHub we’ve figured out the data-driven, standards-based and RDF-native software architecture of a generic platform, which can be used to implement PKG software as well any number of domain-specific use cases in various industries that increasingly employ semantic RDF Knowledge Graphs. Extending and customizing LinkedDataHub’s domain and UI components to implement such use cases is much more efficient compared to building applications from scratch.
结论
Conclusions
本章的标题是对电视剧《奔腾年代》的致敬，旨在传达这样一种观点：知识图谱是一种不可或缺的数据技术，但就其本身而言，它们只是达到目的的手段，而不是目的本身。目的，也就是事物，将是一个更大的东西，即由现代计算技术融合推动的范式转变。
The title of this chapter is a homage to the TV show Halt and Catch Fire paraphrased to convey the idea that Knowledge Graphs are an indispensable data technology but as such, they are only means to an end and not the end itself. The end, the thing, will be something much bigger, a paradigm shift fueled by the convergence of modern computing technologies.
我回顾了个人知识图谱的当前和历史用法，发现当前 PKG 的定义非常狭窄，侧重于记笔记。与“Web 3.0”的情况类似，PKG 的概念已被纳入其本身的一个非常有限的版本。
I have reviewed the current and historical usages of personal Knowledge Graphs and found that the current definition of PKGs is very narrow and focused on note-taking. Similar to what happened with “Web 3.0”, the concept of PKGs has been co-opted into a very limited version of itself.
我也尝试概括 PKG 方法并发现它们具有以下特征：
I have also attempted to generalize the PKG approaches and found that they shall exhibit the following traits:
开放式实体类型系统
Open-ended entity type system
具有标记关系的图形数据模型
Graph data model with labeled relations
以人（用户）为中心实体
Person (the user) presented as the central entity
我已经展示了如何使用 RDF 技术堆栈来建模和实现笔记 PKG 工具的一些最突出的功能（可寻址块和双向链接），使用本体和 SPARQL 查询语言。该模型确实过于简单，因为它不支持将文本跨度注释为内容中的关系，而是将它们建模为单独的三元组。为了启用注释，我们可能需要合并 HTML+RDFa，这是交错 HTML 内容和 RDF 三元组的标准方法。
I have shown how the RDF technology stack can be used to model and implement some of the most prominent features of note-taking PKG tools (addressable blocks and bidirectional links) using ontologies and the SPARQL query language. The model is admittedly simplistic as it does not support annotating spans of text as relations within the content, instead of modeling them as separate triples. In order to enable annotations we would probably need to incorporate HTML+RDFa, which is the standard way to interleave HTML content and RDF triples.
此外，我已经证明了通过定制 LinkedDataHub（一个通用的开源知识图谱平台）已经可以实现通用的 PKG 用例。你不必相信我的话——LinkedDataHub 是一个你可以亲自尝试的开源项目。再一次向《奔腾年代》致敬：请接受它，不是因为它是什么，而是因为它有潜力成为的一切。
Furthermore, I have shown that it is already possible to implement the generalized PKG use case by customizing LinkedDataHub, a generic, open-source Knowledge Graph platform. You don’t have to take my word for it – LinkedDataHub is an open source project that you can try for yourself. And to use another nod to Halt and Catch Fire: please embrace it not for what it is, but for everything it has the potential to be.
附件：语义网标准
Annex: Semantic Web Standards
语义网是 Tim-Berners Lee 提出的愿景，即机器可以直接或间接地解释网络上的数据。语义网的主要思想是创建一个数据网络，其中已识别的实体在网络上链接，并使用机器可读的描述进行语义描述。语义网扩展了他对（传统）网络的愿景，其中文档或网站是相互连接的，这些页面的内容是为人类设计的。
The Semantic Web is the vision, by Tim-Berners Lee, where machines can directly or indirectly interpret data on the web. The main idea of the Semantic Web is to create a web of data, where identified entities are linked on the web and semantically described with machine-readable descriptions. The Semantic Web extends his vision of the (traditional) web, in which documents or websites are connected, and the content of these pages is meant for humans.
语义网的愿景促使万维网联盟 (W3C) 制定了标准，旨在支持 (语义) 网的发展，同时促进互操作性。使用这些标准，许多知识图谱已在网络上发布。值得注意的例子是 DBpedia 和 Wikidata，它们包含来自维基媒体项目的半结构化数据。此外，知识图谱在企业中的使用也越来越多 [ref]，要么为基于开放标准的数据集成提供语义层，要么作为人工智能系统的核心组件。
The vision of the Semantic Web has led to the development of standards by the World Wide Web Consortium (W3C), which aim at supporting the growth of the (semantic) web while promoting interoperability. Using these standards, many knowledge graphs have been published on the web. Notable examples are DBpedia and Wikidata, which contain semi-structured data from Wikimedia projects. In addition, knowledge graphs have been increasingly used in enterprises [ref] either to provide a semantic layer for data integration based on open standards, or as core components of Artificial Intelligence systems.
资源描述框架（RDF）
The Resource Description Framework (RDF)
资源描述框架 (RDF) 是 W3C 推荐的用于在网络上互连数据或声明的模型。RDF 基于 IRI 等网络技术，包括数据模型和用于表示数据的词汇表。在以下小节中，我们将介绍 RDF 的基础知识。
The Resource Description Framework (RDF) is the W3C recommended model for interconnected data or statements on the web. RDF is based on web technologies like IRIs and includes a data model and a vocabulary to represent data. In the following subsections, we introduce the foundations of RDF.
RDF 术语。为了表示语句，RDF 支持三种类型的术语。这些术语在数据建模中用于不同的目的，并且假定相互不相交。下面，我们提供了 RDF 术语的摘要和示例。
RDF Terms. To represent statements, RDF supports three types of terms. These terms serve different purposes in data modelling and are assumed mutually disjoint. In the following, we provide a summary and examples of the RDF terms.
国际化资源标识符 (IRI)：对应于网络上的全局标识符。这些标识符用于明确命名实体。例如，IRIhttps://dbpedia.org/resource/Harry_Potter_(character) 标识 DBpedia 中名为 Harry Potter 的角色。此外，网络上的不同数据源可以使用不同的 IRI 来命名同一实体。例如，Wikidata 中用 https://www.wikidata.org/wiki/Q3244512 表示 Harry Potter 角色。由于 IRI 被视为全局的，因此也可以重用来自其他来源的现有 IRI，而不必创建新的 IRI。
Internationalized Resource Identifiers (IRIs): Correspond to global identifiers on the web. These are used to unequivocally name entities. For example, the IRIhttps://dbpedia.org/resource/Harry_Potter_(character) identifies the character named Harry Potter in DBpedia. Furthermore, different data sources on the web can use different IRIs to name the same entity. For instance, the character Harry Potter is denoted with https://www.wikidata.org/wiki/Q3244512 in Wikidata. As IRIs are assumed global, it is also possible to reuse existing IRIs from other sources instead of creating new ones.
空白节点 (Bnodes)：这些节点允许提及实体而不为其分配名称。这样，bnodes 可以看作存在变量，即它们表示知识图中存在某些东西但没有名称。实际上，bnodes 被分配一个标签，以区分同一知识图中存在的不同 bnodes。与 IRI 不同，bnodes 不被认为是全局的。因此，如果两个不同的图恰好对 bnode 使用相同的标签，则假设这两个 bnodes 指的是同一个实体是不正确的。
Blank nodes (Bnodes): These allow for mentioning entities without assigning them a name. In this way, bnodes can be seen as existential variables, i.e., they denote that something exists in the knowledge graph but does not have a name. In practice, bnodes are assigned a label to distinguish between different bnodes existing in the same knowledge graph. In contrast to IRIs, bnodes are not assumed global. Therefore, if two different graphs coincidentally use the same label for a bnode, it is not correct to assume that the two bnodes are referring to the same entity.
文字：对应于字符串，在知识图谱中用于对简单数据值进行建模，例如纯字符串、数字或日期。例如，哈利波特的全名是“Harry James Potter”。在 RDF 中，文字还可以用数据类型（例如日期、年份、整数、浮点数等）或语言标签（以指示文字是英语、德语等）进行注释。
Literals: Correspond to strings and are used in KGs to model simple data values, e.g., a plain string, a number, or a date. For example, the full name of Harry Potter is “Harry James Potter”. In RDF, literals can in addition be annotated with datatypes (e.g., date, year, integer, float, etc.) or with language tags (to indicate whether the literal is in English, German, etc.).
RDF 三元组和 RDF 图。RDF 包含基于图的数据模型，其中的语句形成有向的标记图。RDF 数据模型的基本单位是 (RDF) 三元组。三元组是形式为 (主语、谓语、宾语) 的元组，其中主语和宾语对应于图的节点。实际上，节点可能表示现实世界或抽象的实体或类，或简单的数据值。
RDF Triples and RDF Graphs. RDF comprises a graph-based data model, where the statements form a directed, labeled graph. The basic unit of the RDF data model is an (RDF) triple. A triple is a tuple of the form (subject, predicate, object) where the subject and object correspond to nodes of a graph. In practice, nodes may represent real-world or abstract entities or classes, or simple data values.
例如，一个节点可以是 Harry Potter，表示具有该名称的实体，另一个节点可以是 Harry James Potter，表示前一个实体的全名，或 Character，表示虚构人物的类别或群体。为了在 RDF 中连接节点，三元组还具有谓词，谓词是一条有向边，用于编码从主语到宾语的关系。例如，三元组可以是以下形式 (Harry Potter，全名，Harry James Potter)。
For example, a node could be Harry Potter that represents an entity with that name, another node could be Harry James Potter to denote the full name of the previous entity, or Character which represents the class or group of fictional characters. To connect nodes in RDF, a triple also has a predicate which is a directed edge that encodes the relationship from the subject to the object. For example, a triple can be of the form (Harry Potter, full name, Harry James Potter).
RDF 三元组是使用 RDF 术语创建的。主语是用 IRI 或 bnode 标识的节点。谓语始终用 IRI 标识。宾语是任何形式的节点，即 IRI、bnode 或文字。在 RDF 中，文字只能出现在三元组的对象位置，因为这些节点不应该具有传出连接或链接。基于此定义，我们可以使用 RDF 术语将上例中的三元组表述如下：
RDF triples are created using RDF terms. The subject is a node identified with an IRI or a bnode. The predicate is always identified with an IRI. The object is a node in any form, i.e., an IRI, bnode, or literal. In RDF, literals can only appear in the object position of triples, as these nodes are not supposed to have outgoing connections or links. Based on this definition, we can formulate the triple from the previous example using RDF terms as follows:
<http://dbpedia.org/resource/Harry_Potter_(角色)>
<http://dbpedia.org/resource/Harry_Potter_(character)>
<http://dbpedia.org/property/fullName> “哈利·詹姆斯·波特”。
<http://dbpedia.org/property/fullName> "Harry James Potter".
（RDF 三元组的示例。主语和谓语对应于 IRI，此示例中的宾语是文字。）
(Example of an RDF triple. The subject and predicate correspond to IRIs, the object in this example is a literal.)
为了简化 IRI 的读写，在某些机器可读格式或 RDF 序列化（例如 Turtle）中，可以通过定义前缀来创建缩写。例如，上面的三元组也可以序列化为：
To ease the reading and writing of IRIs, in some machine-readable formats or seralizations of RDF (e.g., Turtle) it is possible to create abbreviations by defining prefixes. For example, the previous triple can also be serialized as:
@prefix dbr：http://dbpedia.org/resource/。
@prefix dbr: http://dbpedia.org/resource/ .
@prefix dbp:http://dbpedia.org/property/。
@prefix dbp: http://dbpedia.org/property/ .
dbr:Harry_Potter_(角色) dbp:fullName “哈利·詹姆斯·波特”。
dbr:Harry_Potter_(character) dbp:fullName "Harry James Potter" .
（RDF 三元组的示例。主语和谓语对应于 IRI，此示例中的宾语是文字。）
(Example of an RDF triple. The subject and predicate correspond to IRIs, the object in this example is a literal.)
最后，RDF 图是一种知识图谱 [ref]，定义为一组 (RDF) 三元组。请注意，在 RDF 图中，节点始终连接到其他节点（甚至是它们自己），但构造不允许存在孤立节点。
Finally, an RDF graph is a knowledge graph [ref] defined as a set of (RDF) triples. Note that, in RDF graphs, nodes are always connected to other nodes (or even themselves), but isolated nodes are not allowed by construction.
词汇和本体
Vocabularies and Ontologies
虽然 RDF 被视为半结构化数据模型（即并非所有元素都必须用相同的属性来描述），但 RDF 中的词汇表和本体的作用是引入知识图谱中所表示实体的语义或含义。这样，RDF 中的词汇表和本体就为关系、类以及潜在实体提供了标识符。词汇表通常与本体不同，因为它对论域有更“宽松”的定义或约束。
While RDF is considered a semi-structured data model – i.e., not all elements have to be described exactly with the same attributes – the role of vocabularies and ontologies in RDF is to introduce the semantics or meaning of the entities that are represented in the knowledge graph. In this way, vocabularies and ontologies in RDF provide the identifiers for relations, classes, and potentially entities as well. Vocabularies typically differ from ontologies by having more “loose” definitions or constraints over the domain of discourse.
词汇表和本体通常由领域专家团队开发，他们定义主要概念及其之间的关系。实际上，存在可供重复使用的开放本体（例如基因本体）、开放词汇表（例如朋友之友词汇表）以及词汇表目录（例如链接开放词汇表）。
Vocabularies and ontologies are typically developed by teams of domain experts who define the main concepts and relationships between them. In practice, there exist open ontologies (e.g., the Gene Ontology), open vocabularies (e.g., the Friend-of-a-Friend vocabulary), as well as catalogues of vocabularies (e.g. Linked Open Vocabularies) available for reuse.
为了对词汇表和本体进行建模，语义网技术栈还提供了元词汇表和语言。下面，我们将概述 RDF 中定义的词汇表、RDF Schema (RDFS) 以及 Web 本体语言 (OWL) 中更具表现力的构造。
To model vocabularies and ontologies, the Semantic Web technology stack also provides meta-vocabularies and languages. In the following, we present an overview of the vocabulary defined in RDF, the RDF Schema (RDFS), and the more expressive constructs in the Web Ontology Language (OWL).
RDF 和 RDF Schema。除了数据模型之外，RDF 还提供了一个简单的词汇表来表示语句。该词汇表定义在 http://www.w3.org/1999/02/22-rdf-syntax-ns。在本章的其余部分，我们将使用前缀 rdf 来缩写 RDF 词汇表中定义的标识符。
RDF and RDF Schema. In addition to a data model, RDF also provides a simple vocabulary to represent statements. This vocabulary is defined at http://www.w3.org/1999/02/22-rdf-syntax-ns. In the remainder of this chapter, we abbreviate the identifiers defined in the RDF vocabulary with the prefix rdf.
对于本章，RDF 词汇表中最相关的标识符是用于表示 is-a 关系的标识符，表示为 rdf:type。此标识符允许将节点或实体分配给（可能为多个）类。例如，为了表达哈利波特是一个虚构人物，我们可以创建以下 RDF 三元组：
For this chapter, the most relevant identifier in the RDF vocabulary is the one to represent is-a relationships, which is denoted rdf:type. This identifier allows for assigning nodes or entities to (potentially several) classes. For example, to express that Harry Potter is a fictional character, we can create the following RDF triple:
@prefix dbr: <http://dbpedia.org/resource/> 。
@prefix dbr: <http://dbpedia.org/resource/> .
@prefix dbo: <http://dbpedia.org/ontology/> 。
@prefix dbo: <http://dbpedia.org/ontology/> .
@prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#> 。
@prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#> .
dbr：Harry_Potter_（角色）rdf：type dbo：FictionalCharacter。
dbr:Harry_Potter_(character) rdf:type dbo:FictionalCharacter .
图 7.7
Figure 7.7
（在 Turtle 中使用 rdf:type 的 is-a 语句中的 RDF 三元组示例（顶部）和图表（底部））
(Example of an RDF triple with in is-a statement using rdf:type in Turtle (top) and as a diagram (bottom))
为了对 RDF 图的元素进行更具表现力的建模，我们可以使用 RDF Schema (RDFS)。RDFS 词汇表在 http://www.w3.org/2000/01/rdf-schema# 中定义，我们使用前缀 rdfs。RDFS 是 RDF 之上的模型，提供允许定义以下内容的标识符：
To model more expressive statements about the elements of an RDF graph, we can use RDF Schema (RDFS). The RDFS vocabulary is defined at http://www.w3.org/2000/01/rdf-schema# , and we use the prefix rdfs. RDFS is a model on top of RDF and provide identifiers that allow for defining:
类：表示具有相似特征或属性的实体组。类用 IRI rdfs:Class 表示。
Classes: Represent groups of entities with similar characteristics or properties. A class is denoted with the IRI rdfs:Class.
类和属性的层次结构：允许分别表示类和属性之间的泛化/特化关系。对于类，RDFS 提供 IRI rdfs:subClassOf，对于属性，提供 rdfs:subPropertyOf。在 RDF 三元组中，更具体的概念占据主语位置，而更一般的概念出现在宾语位置。
Hierarchies of classes and properties: Allows for representing the relationship of generalization/specialization between classes and properties, respectively. For classes, RDFS provides the IRI rdfs:subClassOf, and for properties rdfs:subPropertyOf. In an RDF triple, the more specific concept takes the subject position, while the more general one occurs in the object position.
资源：在 RDF 中，所有在图中描述的实体都称为资源。因此，RDFS 提供了 IRI rdfs:Resource，它表示一切事物的类。RDF 图中类是 rdfs:Resource 的子类，而实体是它的实例。
Resources: In RDF, all entities that are described in a graph are called resources. Therefore, RDFS provides the IRI rdfs:Resource, which denotes the class of everything. Classes in an RDF graph are then subclasses of rdfs:Resource, while entities are instances of it.
人类可读的信息：尽管 RDF 和其他语义网技术都是针对机器的，但提供人类可读的信息以自然语言描述 KG 的元素也很重要。当 IRI 不包含实际单词而是由代码或数字组成时，这一点更加明显。为了支持这一点，RDFS 提供了两个 IRI：(i) rdfs:label 用于分配人类可读的标签（例如，现实世界中该实体的名称），以及 (ii) rdfs:comment 用于以自然语言提供有关 IRI 的更长描述。在具有任何这些 IRIS 的 RDF 三元组中，对象都是文字。
Human-readable information: Despite the fact that RDF and other Semantic Web technologies are targeted at machines, it is important to offer human-readable information to describe the elements of a KG in natural language. This is more apparent when IRIs do not contain actual words, but are composed of codes or numbers. To support this, RDFS offers two IRIs: (i) rdfs:label to assign a human-readable label (e.g., the name of that entity in the real world), and (ii) rdfs:comment to provide longer descriptions in natural language about the IRI. In RDF triples with any of these IRIS, the objects are literals.
下表总结了本节中介绍的 RDF 和 RDFS 标识符，并说明了这些标识符在 RDF 三元组中的用法。官方文档中提供了完整的 RDFS 定义列表。
The following table summarizes the RDF and RDFS identifiers covered in this section, and illustrates the usage of these identifiers in RDF triples. The full list of RDFS definitions are available at the official documentation.
表 7.1
Table 7.1
Web 本体语言 (OWL)。OWL 是一种本体语言，用于表示有关实体、类和属性的更高级语句（与 RDF/S 相比）。如果本体是使用 OWL 中定义的概念构建的，则称为 OWL 本体。OWL 通常用于建模需要高度表达性语句的领域，这些语句涉及实体在现实世界中如何关联，这些语句超越了类/属性层次结构（如 RDFS）。
The Web Ontology Language (OWL). OWL is an ontology language to represent more advanced statements (in comparison to RDF/S) about entities, classes, and properties. An ontology is called an OWL ontology if it is constructed using the concepts defined in OWL. OWL is typically used for modelling domains that require highly expressive statements about how entities are related in the real-world that go beyond class/property hierarchies (as in RDFS).
下面，我们概述了 OWL 在实体、类和属性级别定义的构造或公理类型。OWL 定义在 http://www.w3.org/2002/07/owl# 中，我们使用前缀 owl。OWL 定义的完整列表可在官方文档中找到。
In the following, we provide an overview of the type of constructs or axioms that are defined in OWL at the level of entities, classes, and properties. OWL is defined at http://www.w3.org/2002/07/owl# and we use the prefix owl. The full list of OWL definitions are available at the official documentation.
个别公理：OWL提供了两个谓词来定义实体之间的关系，owl:sameAs表示两个IRI在论域中代表同一个实体，owl:differentFrom明确表达它们并不相同。
Individual axioms: OWL provides two predicates to define the relationship between entities. owl:sameAs indicates that two IRIs represent the same entity in the domain of discourse, and owl:differentFrom to explicitly express that they are not the same.
类公理：这些公理包含比 RDFS 更具表现力的类间关系。例如，在 OWL 中，可以使用谓词 owl:equivalentClass 表示两个类具有完全相同的实体，或者使用谓词 owl:disjointWith 表示它们不共享任何实体。
Class axioms: These axioms comprise more expressive relationships between classes than RDFS. For example, in OWL it is possible to express that two classes have the exact same entities using the predicate owl:equivalentClass or that they do not share any entities using the predicate owl:disjointWith.
属性公理：这些公理提供了比 RDFS 更具表现力的谓词关系。例如，在 OWL 中，可以使用谓词 owl:inverseOf 来表示两个关系互为逆，或者通过将属性分配给类 owl:TransitiveProperty 来表示属性是传递性的。
Property axioms: These axioms provide more expressive relationships between predicates than RDFS. For example, in OWL it is possible to express that two relationships are the inverse of each other using the predicate owl:inverseOf or that a property is transitive by assigning it to the class owl:TransitiveProperty.
属性定义：除了属性公理之外，OWL 还根据三元组的对象值区分两种类型的属性。owl:DatatypeProperty 将主体与文字对象值相关联；否则，属性被定义为 owl:ObjectProperty。
Property definitions: In addition to property axioms, OWL distinguishes between two types of properties depending on the object value of a triple. An owl:DatatypeProperty relates a subject to a literal object value; otherwise, the property is defined as an owl:ObjectProperty.
基数限制：使用 OWL，可以定义谓词基数的限制或约束，即具有给定谓词的主体可以具有的值的数量。这些限制可以用 owl:minCardinality、owl:maxCardinality 和 owl:exactCardinality 来定义。
Cardinality Restrictions: With OWL, it is possible to define restrictions or constraints over the cardinalities of predicates, i.e., the number of values that a subject with a given predicate can have. These restrictions can be defined with owl:minCardinality, owl:maxCardinality, and owl:exactCardinality.
OWL 公理在知识图谱上的定义允许推断有关数据的新事实；这些事实通常被称为隐性知识，当将公理与形式上定义 OWL 中定义的构造含义的蕴涵规则结合使用时，这些事实可以具体化（或显化）。同样，这些公理和限制可用于识别知识图谱中的不一致之处。
The definition of OWL axioms over KGs allows for inferring new facts about the data; these facts are usually referred to as implicit knowledge that can be materialized (or made explicit) when applying the axioms in combination with the entailment rules that formally define the meaning of the constructs defined in OWL. Likewise, these axioms and restrictions can be applied to identify inconsistencies in the KG.
查询 RDF 图：SPARQL 查询语言
Querying RDF Graphs: The SPARQL Query Language
W3C 推荐的查询 RDF 图的语言是 SPARQL 协议和查询语言 (SPARQL)。SPARQL 是一种声明性语言，其中查询使用图模式指定。下面，我们介绍 SPARQL 支持的不同结构。
The W3C recommended language to query RDF graphs is the SPARQL Protocol and Query Language (SPARQL). SPARQL is a declarative language, where queries are specified with graph patterns. In the following, we present an introduction to the different structures supported in SPARQL.
三元组模式。SPARQL 中定义的原子单位是三元组模式，它类似于 RDF 三元组。在三元组模式中，主语、谓语或宾语（包括同时有多个）可以用占位符或变量替换。变量以 $ 或 ? 为前缀。例如，要在 DBpedia 中询问哈利波特角色的创作者，我们可以编写以下三元组模式。
Triple Patterns. The atomic unit defined in SPARQL is a triple pattern, which is similar to an RDF triple. In a triple pattern, the subject, predicate, or object (including several at the same time) can be replaced by a placeholder or variable. Variables are prefixed by $ or ?. For example, to ask for the creator of the character Harry Potter in DBpedia, we can write the following triple pattern.
<http://dbpedia.org/resource/Harry_Potter_(角色)>
<http://dbpedia.org/resource/Harry_Potter_(character)>
<http://dbpedia.org/ontology/creator> ?o 。
<http://dbpedia.org/ontology/creator> ?o .
（查询角色“哈利波特”的创作者的三元组模式示例。变量 ?o 表示我们正在请求 RDF 三元组中与给定主语和谓语匹配的对象值）
(Example of a triple pattern to query the creator of the character Harry Potter. The variable ?o indicates that we are requesting for the object values in RDF triples that match the given subject and predicate)
基本图形模式 (BGP)。BGP 由一组三重模式组成。BGP 表示应在 RDF 图上匹配的条件的结合。例如，以下 BGP 检索类 Fictional Character 及其全名的所有实例。值得注意的是，在此示例中，变量 ?s 在两个三重模式中均有使用，以指示一个三重模式的匹配也应用于获取另一个三重模式的答案。通过这种方式，SPARQL 表示其他查询语言（如 SQL）中已知的连接。
Basic Graph Patterns (BGPs). A BGP is composed of a set of triple patterns. BGPs represent conjunctions of conditions that should be matched over the RDF graph. For example, the following BGP retrieves all the instances of the class Fictional Character and their full name. It is important to note that, in this example, the variable ?s is used in both triple patterns to indicate that the matches of one triple pattern should also be used to get the answers of the other triple pattern. In this way, SPARQL represents joins as known in other query languages like SQL.
?s <http://www.w3.org/1999/02/22-rdf-syntax-ns#type> <http://dbpedia.org/ontology/FictionalCharacter> 。
?s <http://www.w3.org/1999/02/22-rdf-syntax-ns#type> <http://dbpedia.org/ontology/FictionalCharacter> .
?s <http://dbpedia.org/property/fullName> ?名称 。
?s <http://dbpedia.org/property/fullName> ?name .
（基本图形模式 (BGP) 的示例。第一个三元组模式检索虚构人物类的所有实体（存储在变量 ?s 中）。对于这些角色，检索全名（存储在变量 ?name 中））
(Example of Basic Graph Pattern (BGP). The first triple pattern retrieves all the entities (stored in the variable ?s) of the class of Fictional Characters. For those characters, retrieve the full name (stored in the variable ?name))
图形模式
Graph Patterns
除了 BGP 之外，SPARQL 还支持其他类型的表达式来表示 BGP 或其他 SPARQL 表达式的析取（使用关键字 UNION）、BGP 或其他 SPARQL 表达式的可选匹配（使用关键字 OPTIONAL）以及过滤条件（使用关键字 FILTER）。
Besides BGPs, SPARQL supports further types of expressions to represent disjunctions of BGPs or other SPARQL expressions (with the keyword UNION), optional matches of BGPs or other SPARQL expressions (with the keyword OPTIONAL), and filtering conditions (with the keyword FILTER).
{?s <http://www.w3.org/1999/02/22-rdf-syntax-ns#type> <http://dbpedia.org/ontology/FictionalCharacter> . }
{?s <http://www.w3.org/1999/02/22-rdf-syntax-ns#type> <http://dbpedia.org/ontology/FictionalCharacter> . }
{?s <http://dbpedia.org/property/fullName> ?名称 .}
{?s <http://dbpedia.org/property/fullName> ?name .}
联盟
UNION
{?s <http://dbpedia.org/property/nicknames> ?昵称 .}
{?s <http://dbpedia.org/property/nicknames> ?nickname .}
（使用 UNION 的图形模式示例。第一个三重模式检索虚构人物类的所有实体（存储在变量 ?s 中）。对于这些人物，检索他们的全名（存储在变量 ?name 中）或他们的昵称（存储在变量 ?nickname 中）。如果主题具有两个属性的值，则返回两个值。
(Example of a graph pattern using UNION. The first triple pattern retrieves all the entities (stored in the variable ?s) of the class of Fictional Characters. For those characters, retrieve their full name (stored in the variable ?name) or their nickname (stored in the variable ?nickname). If a subject has values for both properties, both values are returned.
SPARQL 选择查询
SPARQL Select Queries
SELECT 查询遵循 SELECT-FROM-WHERE 结构。关键字 SELECT 后跟最终答案的变量子集；快捷方式 * 允许检索查询的 WHERE 部分中定义的所有变量。关键字 FROM 用于指定查询的 RDF 图的 IRI。根据用于存储 RDF 图的后端，可以省略查询的 FROM 部分。关键字 WHERE 是查询的主体，其中匹配条件在 { } 之间使用三元组模式、BGP 或其他图形模式指定。
A SELECT query follows the structure SELECT-FROM-WHERE. The keyword SELECT is followed by a subset of variables that are part of the final answer; the shortcut * allows for retrieving all variables defined in the WHERE part of the query. The keyword FROM is used to specify the IRI of an RDF graph that is queried. Depending on the backend that is used to store the RDF graph, the FROM part of the query can be omitted. The keyword WHERE is the body of the query, where the matching conditions are specified between { } using triple patterns, BGPs, or other graph patterns.
选择？全名？创作者
SELECT ?fullname ?creator
在哪里 {
WHERE {
{?s <http://www.w3.org/1999/02/22-rdf-syntax-ns#type> <http://dbpedia.org/ontology/FictionalCharacter> 。
{?s <http://www.w3.org/1999/02/22-rdf-syntax-ns#type> <http://dbpedia.org/ontology/FictionalCharacter> .
?s <http://dbpedia.org/property/fullName> ?名称 .}
?s <http://dbpedia.org/property/fullName> ?name .}
选修的
OPTIONAL
{?s <http://dbpedia.org/ontology/creator> ?创作者 .}
{?s <http://dbpedia.org/ontology/creator> ?creator .}
}
}
（SPARQL 选择查询的示例。只有变量 ?fullname 和 ?creator 是答案的一部分。WHERE 部分由 BGP 和三元组模式之间的 OPTIONAL 表达式组成）
(Example of a SPARQL select query. Only variables ?fullname and ?creator are part of the answer. The WHERE part is composed of an OPTIONAL expression between a BGP and a triple pattern)
SPARQL 更新查询。SPARQL 1.1 提供了一种更新语言来修改 RDF 图的内容。图更新操作由关键字 INSERT DATA 定义，用于添加三元组，DELETE DATA 定义，用于删除三元组，DELETE/INSERT 允许修改 RDF 图中与某些 SPARQL 表达式匹配的所有三元组。
SPARQL Update Queries. SPARQL 1.1 provides an update language to modify the content of an RDF graph. The graph update operations are defined by the keywords INSERT DATA to add triples, DELETE DATA to remove triples, and DELETE/INSERT allows for modifying all the triples in the RDF graph that match some SPARQL expressions.
INSERT DATA 和 DELETE DATA 操作处理 RDF 三元组的创建和删除，即不允许使用三元组模式。此外，我们可以通过添加关键字 GRAPH 来指定操作所在的 RDF 图的 IRI。下面显示了在 SPARQL 中创建三元组的示例。
The INSERT DATA and DELETE DATA operations handle the creation and deletion of RDF triples, i.e., no triple patterns are allowed. In addition, we can specify the IRI of the RDF graph where the operation is taking place by adding the keyword GRAPH. An example for creating a triple in SPARQL is shown below.
插入数据 {
INSERT DATA {
{ 图表 <http://dbpedia.org/> { dbr:Harry_Potter_(character) dbo:spouse dbr:Ginny_Weasley . }}
{ GRAPH <http://dbpedia.org/> { dbr:Harry_Potter_(character) dbo:spouse dbr:Ginny_Weasley . }}
（使用 INSERT DATA 将 RDF 三元组添加到使用 IRI http://dbpedia.org/ 标识的图表中。DELETE DATA 操作遵循相同的语法）
(Adding an RDF triple to the graph identified with the IRI http://dbpedia.org/ using INSERT DATA. The operation DELETE DATA follows the same syntax)
DELETE/INSERT 操作是操作 RDF 图内容的更通用方法。此操作的完整语法形式为 WITH-DELETE/INSERT-WHERE。WITH 是可选的，其行为类似于 SPARQL Select 查询中的关键字 FROM，即，它指定将在哪个 RDF 图上执行后续操作。然后，表达式后面跟着 DELETE 以模拟删除、INSERT 以模拟插入或两者以模拟更新。
The DELETE/INSERT operation is a more general way of manipulating the content of an RDF graph. The full syntax of this operation is of the form WITH-DELETE/INSERT-WHERE. WITH is optional and behaves similarly to the keyword FROM in SPARQL Select queries, i.e., it specifies over which RDF graph the subsequent operations will be performed. Then, the expression is followed by a DELETE to model deletions, an INSERT to model insertions, or both to model updates.
这些表达式可以包含与 WHERE 子句的结果绑定的 BPG。最后，在 WHERE 子句中，我们可以指定应匹配的条件（使用 SPARQL 表达式）以执行相应的插入、删除或更新。以下是如何使用 DELETE/INSERT 操作的示例。
These expressions can contain BPGs which are bound with the results from the WHERE clause. Lastly, in the WHERE clause, we can specify the conditions that should be matched (using SPARQL expressions) to perform the corresponding insertions, deletions, or updates. The following are examples how to use the DELETE/INSERT operation.
与 <http://dbpedia.org/>
WITH <http://dbpedia.org/>
删除 { dbr:Harry_Potter_(character) dbp:nicknames "哈利·波特" . }
DELETE { dbr:Harry_Potter_(character) dbp:nicknames "Harry Potter" . }
插入 { dbr:Harry_Potter_(character) dbp:nicknames “幸存的男孩” . }
INSERT { dbr:Harry_Potter_(character) dbp:nicknames "The Boy Who Lived" . }
（更新 RDF 三元组。在用 IRI http://dbpedia.org/ 标识的 RDF 图中，删除 DELETE 子句中指示的昵称的三元组，并添加 INSERT 子句中指示的另一个昵称的新三元组。这也相当于使用 DELETE DATA 和 INSERT DATA 各与 GRAPH 关键字一起使用）
(Update of an RDF triple. In the RDF graph identified with the IRI http://dbpedia.org/, delete the triple with the nickname indicated in the DELETE clause, and add a new triple with another nickname indicated in the INSERT clause. This is also equivalent to use DELETE DATA and INSERT DATA each with the GRAPH keyword)
与 <http://dbpedia.org/>
WITH <http://dbpedia.org/>
删除 { ?s dbo:spouse ?x . }
DELETE { ?s dbo:spouse ?x . }
插入 { ?s rdf:type dbo:FictionalCharacter .
INSERT { ?s rdf:type dbo:FictionalCharacter .
?s dbo:配偶 ?x . }
?s dbo:spouse ?x . }
（删除几个 RDF 三元组。在使用 IRI http://dbpedia.org/ 标识的 RDF 图中，删除所有出现的关系 dbo:spouse（在 DELETE 子句中指示），对于所有有配偶的 dbo:FictionalCharacter 类实体（在 WHERE 子句中指示））
(Deletion of several RDF triples. In the RDF graph identified with the IRI http://dbpedia.org/, delete all occurrences of the relationship dbo:spouse (indicated in the DELETE clause), for all entities of the class dbo:FictionalCharacter that have a spouse (indicated in the WHERE clause))
存储 RDF 图
Storing RDF Graphs
RDF 图可以使用 RDF 格式序列化为机器可读的文档。我们之前在本节中已经看到了一种格式的示例（即 Turtle）。这些 RDF 文档可以加载到主内存中，并由应用程序使用不同编程语言中提供的专用库直接操作。下表简要列出了可用于处理 Python、JavaScript、Java 和 C 中的 RDF 图的开源库和存储库。
RDF graphs can be serialized into machine-readable documents using formats for RDF. We already saw an example of one format (i.e., Turtle) previously in this section. These RDF documents can be loaded in main memory and directly manipulated by applications using specialized libraries available in different programming languages. The following table presents a brief list of available open-source libraries and repositories for handling RDF graphs in Python, JavaScript, Java, and C.
语言
Language
RDF 库示例
Example of RDF library
Python
Python
JavaScript
JavaScript
Java
Java
碳
C
当 RDF 图变得非常大时，完全使用内存库来处理它们就不再实用了。相反，RDF 图可以存储在数据库系统中，这些数据库系统为加载和查询大量数据提供更高级的支持。原生支持 RDF 的系统称为三元组存储。RDF 图也可以用其他技术来处理，例如关系数据库管理系统 (RDBMS) 甚至 NoSQL 存储。这些 RDBMS 提供适当的接口来支持 SPARQL 查询。
When RDF graphs become very large, it is not practical anymore to handle them with in-memory libraries entirely. Instead, RDF graphs can be stored in database systems that provide more advanced support for loading and querying large amounts of data. Systems that natively support RDF are called triplestores. RDF graphs can also be handled with other technologies like Relational Database Management Systems (RDBMS) and even NoSQL stores. These RDBMS provide appropriate interfaces to support SPARQL queries.
第八章
Chapter 8
利用语义网标准构建个人知识图谱
Leveraging Semantic Web Standards for Personal Knowledge Graphs
Maribel Acosta 和 Omes Baltes
介绍
Introduction
数据和信息的不断增长超出了人类大脑的自然能力。这种现象被称为信息过载问题。个人知识图谱 (PKG) 近来引起了业界越来越多的关注，被认为是解决个人信息过载问题的有希望的解决方案。人们早已认识到个人构建和组织知识的必要性。在过去的几十年里，人们开发了大量的生产力工具来增强人类思维的能力。文本处理、文件和任务管理、词汇训练、笔记记录和思维导图等应用程序都试图帮助人们管理他们掌握的信息和知识。
The never-ending growth of data and information is overwhelming the natural capacities of the human brain. This phenomenon is known as the information overload problem. Personal knowledge graphs (PKGs) have recently garnered increasing attention from the industry as a promising solution to information overload on the individual level. The need for individuals to structure and organize knowledge has long been recognised. Over the last decades, a multitude of productivity tools have been developed to augment the abilities of the human mind. Applications for text processing, file and task management, vocabulary training, note-taking, and mind-mapping, all try to help people manage the information and knowledge at their disposal.
近年来，出现了一波个人知识管理工具，专注于让所有这些知识都可访问且相互关联的需求。这种自由联想的知识方法应该模仿人类的思维方式。其中一些工具以此为动机，将自己推销为“思维工具”，声称能够维持终身的“第二大脑”、“个人知识图谱”或“数字花园”。然而，即使是今天的个人知识管理工具也不足以解决信息过载问题。更糟糕的是，这些工具存在一些局限性，使其不适合终身个人知识管理，包括：
In more recent years, there has been a wave of personal knowledge management tools focusing on the need to have all of this knowledge accessible and interlinked. This freely associative approach to knowledge is supposed to mimic the way humans think. Some of these tools take this as motivation to market themselves as “tools for thought”, claiming to enable maintenance of a lifelong “second brain”, “personal knowledge graph”, or “digital garden”. Still, even today’s tools for personal knowledge management are not powerful enough to address information overload. Even worse, these tools have limitations that make them unfit candidates for lifelong personal knowledge management, including:
专有技术而非开放标准。它们基于专有技术，与其他数据格式和服务的互操作性较差。这导致供应商锁定工具，这意味着 PKG 中的数据无法在其他工具中使用，除非付出相当大的转换成本。用户的数据有时存储在数据孤岛中，这意味着他们无法控制和拥有自己的数据。
Proprietary technology instead of open standards. They are based on proprietary technology and have poor interoperability with other data formats and services. This results in vendor lock-in tools, which means the data within the PKG cannot be used in other tools without considerable switching costs. The users’ data is sometimes stored in data silos, meaning they do not have control and ownership over their data.
缺乏结构和表达能力。它们大多基于非结构化的纯文本，不支持结构化知识、模式或语义。这限制了 PKG 工具的某些功能。此外，虽然一些 PKG 工具支持用户定义的模式，但这些模式并非基于标准。这意味着与外部知识数据库集成或协作是一项相当困难的任务。但事实证明，这些特征对于行业和学术界的知识管理是必不可少的。
Lack of structure and expressivity. They are mostly based on unstructured plain text and do not support structured knowledge, schema or semantics. This limits some of the functionalities of PKGs tools. In addition, while a few PKG tools support user-defined schemas, these are not based on standards. This means integrating or collaborating with external knowledge databases is a rather difficult task. But these characteristics have proven indispensable for knowledge management in industry and academics.
缺少查询语言支持。这是前几点的直接后果。只有结构化信息（如存储在数据库中的数据）才支持查询语言。这些工具大多不支持查询语言，而是依赖于自己的搜索算法。这将它们限制为单一的信息检索路径：全文搜索。
Missing query language support. A direct consequence of the previous points. Only structured information, like data stored in databases, supports query languages. These tools mostly do not support query languages and rely on rolling their own search algorithms. This limits them to a single information retrieval path: full-text search.
为了解决上述问题和限制，我们需要研究可用于部署知识图谱的现有技术。特别是，万维网联盟 (W3C) [1] 在语义网技术的保护下制定了标准，为在网络上部署知识图谱提供了理论和实践基础。这些标准包括提供基于图形的数据模型的资源描述框架 (RDF)，以及 RDF 模式 (RDFS) 和 Web 本体语言 (OWL)，它们增加了进一步的语义表达能力以表示知识图谱中的语句。查询语言标准存在于 SPARQL 协议和 RDF 查询语言 (SPARQL) 中。鉴于这些基础，在本章中，我们将探讨使用语义网标准 (SWS) 构建知识图谱，并讨论在采用这些标准构建知识图谱时应考虑的一些技术和非技术方面。
To address the problems and limitations outlined, we need to look into existing technologies that can be used to deploy PKGs. In particular, the World Wide Web Consortium (W3C) [1] have developed standards, under the umbrella of Semantic Web technologies, that provide the theoretical and practical foundations for deploying KGs on the web. These standards include the Resource Description Framework (RDF) that provides a graph-based data model, as well as RDF Schema (RDFS) and the Web Ontology Language (OWL) that add further semantic expressivity to represent statements in the KG. A query Language standard exists in the SPARQL Protocol and RDF Query Language (SPARQL). Given these foundations, in this chapter, we explore the construction of PKGs using Semantic Web Standards (SWS) and discuss some of the technical and non-technical aspects that should be considered when adopting these standards for PKGs.
我们的方法
Our Approach
在我们的解决方案中，我们采用了使用开放语义网标准构建的 PKG 的愿景，以提供不同用户所需的灵活性和表现力。标准使产品之间的互操作性和重用性成为可能。这在谈论个人数据和知识时至关重要。标准化文件系统或连接器（USB-C）使人们能够轻松共享和连接他们的设备、连接器和数据。然而，数据也需要标准化，因为专有数据不能轻易交换。这会阻碍集成和协作等流程。人们应该能够共享他们的数据和知识，并独立于他们使用的硬件和软件来处理它们。标准化数据格式的很好例子是 iCalendar 和 vCards。它们使人们能够使用和共享联系人和日历事件，无论他们使用哪种设备和应用程序。基于标准的 PKG 将使人们能够对他们的所有个人数据和笔记做同样的事情。然而，期望非技术用户使用 RDF 等高科技标准来管理他们的个人知识是不现实的。因此，在我们提出的解决方案中，我们首先提出一个基于 RDF、RDFS 和 OWL 的语义模型，该模型对后端 PKG 中的含义、结构和内容进行编码。
In our solution, we embrace the vision of a PKG that is built using open Semantic Web Standards to provide flexibility and expressivity as required by different users. Standards enable interoperability and reuse between products. This is crucial when talking about personal data and knowledge. Standardized file systems or connectors (USB-C) enable people to share and connect their devices, connectors and data easily. The data however also needs to be standardized, as proprietary data can not easily be interchanged. This hinders processes like integration and collaboration. People should be able to share their data and knowledge, and work with it independently of which hard- and software they use. Good examples of standardized data formats are iCalendar and vCards. They enable people to use and share contacts and calendar events regardless of which devices and applications they use. Standards based PKG would enable people to do the same with all their personal data and notes. However, expecting non-technical users to manage their personal knowledge with high-tech standards like RDF is unrealistic. Therefore, in our proposed solution, we present first a semantic model based on RDF, RDFS, and OWL that encodes the meaning, structure, and content in the PKG in the backend.
然后，我们方法中的表示层能够解释模型以呈现底层 PKG。通过这种方式，我们的解决方案抽象出了处理语义网标准的技术挑战（非专家用户），同时受益于为语义网开发的所有功能，即能够表达语义、使用外部模式、推理，并支持成熟的查询语言（针对专家用户）。最后，我们以交互式 Web 应用程序的形式开发数据模型的原型实现。
Then, a presentation layer in our approach is able to interpret the model to render the underlying PKG. In this way, our solution abstracts away the technical challenges of dealing with Semantic Web Standards for (non-expert) users, while benefiting from all the features developed for the Semantic Web, i.e., being able to express semantics, use external schema, reasoning, and have support for mature query languages (for expert users). Finally, we develop a prototypical implementation of the data model in the form of an interactive web application.
章节结构
Chapter Structure
本章的其余部分结构如下。在“个人知识图谱的语义网标准”中，我们鼓励使用 SWS 来部署 PKG，并评估它们如何克服本介绍中前面提到的一些限制。接下来，我们介绍我们方法的第一部分，即“个人知识图谱的语义模型”，并解释如何使用 SWS 表示 PKG 的元素。
The remainder for this chapter is structured as follows. In “Semantic Web Standards for Personal Knowledge Graphs” we motivate the application of SWS for deploying PKGs and evaluate how they can overcome some of the limitations previously mentioned in this introduction. Next, we present the first part of our approach, namely the “Semantic Model for Personal Knowledge Graphs” and explain how the elements of PKGs can be represented using SWS.
接下来是我们方法的第二部分，即“表示层”；本节展示了如何在 PKG 的语义数据之上构建 PKG 用户界面。然后使用“原型”说明我们的方法。最后，在“结论和展望”中，我们总结了我们的方法并介绍了未来的工作方向。
This is followed by the second part of our approach, i.e., the “Presentation Layer”; this section shows how the PKG user interface is built on top of the semantic data of the PKG. Our approach is then illustrated with a “Prototype”. Lastly, in “Conclusion and Outlook”, we summarize our approach and present future lines of work.
个人知识图谱的语义网标准
Semantic Web Standards for Personal Knowledge Graphs
语义网标准 (SWS) 包括一套开放标准技术栈，用于在网络上发布和连接机器可读数据。SWS 的目标是让计算机能够使用网络上的语义数据完成“更有用的工作” [2]。这种数据通常被称为自描述数据，这意味着数据通过使用众所周知的词汇表或本体对实体进行语义描述来编码其含义。
Semantic Web Standards (SWS) comprise a technology stack of open standards for publishing and connecting machine-readable data on the web. The goal of SWS is to provide computers the capabilities of doing “more useful work” [2] using semantic data on the Web. This data is typically known as self-describing data, which means that the data encodes its meaning via semantic descriptions of entities using well-known vocabularies or ontologies.
除了语义之外，SWS 的另一个重要方面是实体之间的链接概念，即数据通过具有定义含义的关系连接起来。下面，我们简要讨论 SWS 的特点，这些特点使它们非常适合管理（语义）PKG。
Besides semantics, another important aspect of SWS is the notion of links between entities, i.e., the data is connected via relations with a defined meaning. In the following, we briefly discuss the features of SWS that make them a very good fit for managing (semantic) PKGs.
对于不熟悉语义网标准或需要复习的读者，我们建议他们参阅本章末尾提供的附件“图不是事物本身，而是让我们找到事物的东西”。[3] 本附件提供了有关资源描述框架 (RDF)、词汇表和本体以及存储和查询 RDF 图的信息。
We refer readers not familiar with Semantic Web standards, or in need of a refresher, to the Annex provided at the end of the chapter “Graphs are not the thing, they're the thing that gets us to the thing”. [3] This Annex provides information on the Resource Description Framework (RDF), vocabularies and ontologies, as well as storing and querying RDF graphs.
开放标准。SWS 作为开放标准，提供底层数据的可移植性和互操作性。这对于 PKG 至关重要，因为它能够集成和提取来自不同来源的知识。
Open standards. SWS as open standards provide portability and interoperability of the underlying data. This is essential for a PKG because it enables integrating and extracting knowledge from diverse sources.
可访问性。SWS 建立在为访问数据奠定技术基础的 Web 技术之上。PKG 中的所有数据都需要可访问。SWS 通过将数据存储（即集中式、远程式、逻辑上划分为多个图表等）从使用数据的应用程序抽象出来，从而实现数据独立性。SWS 的数据独立性是通过通用数据模型（即 RDF）和访问机制（使用 IRI 解引用或 SPARQL 协议）实现的。
Accessibility. SWS are built upon web technologies that lay the technical foundations for accessing data. All data in a PKG needs to be accessible. SWS enable data independence by abstracting away the storage of the data – i.e., centralised, remote, logically partitioned in several graphs, etc. – from the application that consumes the data. Data independence with SWS is achieved by common data models (i.e. RDF) and access mechanisms (with IRI dereferencing or the SPARQL protocol).
语义。PKG 可以具有不同的内容形式：结构化、半结构化和非结构化。为了向内容添加语义并创建知识，SWS 提供基于逻辑的形式（元）词汇和本体来描述实体的含义。因此，用户可以根据需要基于 SWS 形式主义开发自己的模式（尽可能富有表现力/复杂），或者在其 PKG 中重复使用现有词汇，从而为其 PKG 增添含义。
Semantics. A PKG can have different forms of content: structured, semi-structured, and unstructured. To add semantics to the content and create knowledge, SWSs provide formal (meta-)vocabularies and ontologies based on logic to describe the meaning of entities. Therefore, users can include meaning to their PKGs by developing their own schemata based on SWS formalisms as expressive/complex as necessary, or by reusing existing vocabularies in their PKGs.
嵌入。嵌入也称为嵌入或组合，它描述了相同数据出现在多个位置的能力。这是通过引用包含数据而不是复制数据来实现的。例如，嵌入使 PKG 中使用相同内容（例如段落）的多个页面能够在任何位置的内容发生变化时立即更新。由于 SWS 基于 Web 技术，因此链接和引用数据的概念本身就受到支持。
Transclusion. Also known as embedding or composition, transclusion describes the ability for the same data to appear in several places. This is enabled by including data by reference, rather than duplicating it. For example, transclusion enables multiple pages in a PKG that use the same piece of content (for example, a paragraph) to be updated as soon as the content is changed in any location. Since SWS are based on web technologies, the concept of linking and referencing data is natively supported.
我们的方法：概述
Our Approach: Overview
在本章中，我们介绍了一种使用语义网标准 (SWS) 实现语义 PKG 的方法。我们解决方案的核心是一个语义模型，它包括用语义技术（即 RDF、RDFS 和 OWL）表示的 PKG 中包含的知识。对于 PKG 中的不同元素，语义模型包括几个具有专用目的的数据层。语义模型在“个人知识图谱的语义模型”一节中描述。在此模型之上，我们定义了一个表示层，该层以用户友好的界面呈现语义模型中的知识。表示层显示在“表示层”一节中。语义模型和表示层之间的通信可以通过 SPARQL 查询执行，以更新或检索要显示在用户界面中的数据层内容。图 8.3 描绘了我们的解决方案的草图。
In this chapter, we present an approach for semantic PKGs using Semantic Web Standards (SWS). At the core of our solution is a semantic model, which includes the knowledge contained in the PKG represented with semantic technologies, i.e., RDF, RDFS and OWL. For different elements in the PKG, the semantic model includes several data layers with dedicated purposes. The semantic model is described in the section “A Semantic Model for Personal Knowledge Graphs”. On top of this model, we define a presentation layer that renders the knowledge in the semantic model in a user-friendly interface. The presentation layer is shown in the section “Presentation Layer”. The communication between the semantic model and the presentation layer can be executed via SPARQL queries to update or retrieve content from the data layers to be displayed in the user interface. Figure 8.3 depicts a sketch of our solution.
图 8.1
Figure 8.1
我们提出的方法概述。语义层包含使用 SWS 在 PKG 中表示的所有数据。表示层呈现底层 RDF 图以生成用户友好的界面
Overview of our proposed approach. A semantic layer contains all the data represented in the PKG using SWS. The presentation layer renders the underlying RDF graph to generate user-friendly interfaces
个人知识图谱的语义模型
A Semantic Model for Personal Knowledge Graphs
如上一节所述，我们需要一个灵活而强大的数据模型，能够表达原子和复杂概念，并在 PKG 中自由关联它们。从现在开始，我们将用模型表示的实体或元素称为知识元素。通常，在 PKG 中，我们可以区分不同类型的知识元素，主要是：
As outlined in the previous section, we require a data model that is flexible and powerful, able to express both atomic and complex concepts and freely associate them in a PKG. From now on, we are going to refer to the entities or elements that can be represented with a model as knowledge elements. Typically, in PKGs, we can distinguish between different types of knowledge elements, mainly:
实体（节点）用于识别其描述的领域中的概念。当前的 PKG 工具使用段落、块或页面作为实体。
Entities (nodes)are used to identify concepts from the domain they are describing. Current PKG tools use either paragraphs, blocks or pages as entities.
实体或属性之间的关系（边、弧、链接）用于添加连接、元数据和语义。这些关系构成了实际的图形。当前的工具缺乏关系语义，并且没有方向。它们也没有 OWL 中的传递或功能关系等特征。
Relationships (edges, arcs, links) between entities or attributes to add connections, metadata and semantics. These relationships are what spans the actual graph. Current tools lack semantics on relationships and are not directed. They also don’t have characterizations like transitive or functional relationships found in OWL.
属性（原子数据），用于将基本信息附加到实体。许多当前的 PKG 工具不支持此功能。
Attributes (atomic data), to attach fundamental information to the entity. This feature is not supported in many current PKG tools.
以块形式呈现的文本。这些块包含或嵌套在实体（页面）中，并提供与页面相关的文本描述或信息。
Text in the forms of blocks. These blocks are contained or nested within entities (pages) and provide textual descriptions or information related to the page.
这些出现在 PKG 中的知识元素可以使用 RDF 数据模型进行存储。为此，我们将它们映射到 RDF 中的元素。表 8.1 显示了这些映射。
These knowledge elements that occur in PKGs can be stored using the RDF data model. For this, we map them to elements in RDF. The table 8.1 shows these mappings.
表 8.1
Table 8.1
为了存储前面描述的知识元素，我们提出了一个基于 RDF 的 PKG 管理语义模型，其中数据分为不同的层：
To store the previously described knowledge elements, we propose a semantic model for managing a PKG based on RDF, where the data is organized in different layers:
事实层。包含（半）结构化和非结构化知识。这是添加到 PKG 的所有数据的主要存储。此层中的数据以 RDF 三元组的形式保存。
Facts Layer. Contains (semi-)structured and unstructured knowledge. This is the main storage of all the data added to the PKG. Data in this layer is persisted as RDF triples.
架构层。此层默认包含基本架构，如 RDF、RDFS 和 OWL。此外，用户可以创建自己的个人架构，并编辑 PKG 的类、关系和属性。此层中的数据以 RDF 三元组的形式保存。
Schema Layer. This layer includes by default basic schemas like RDF, RDFS and OWL. In addition, the user can create their own personal schema, and edit classes, relationships, and attributes of the PKG. Data in this layer is persisted as RDF triples.
临时数据层。用于存储外部 RDF 数据和从非结构化文本知识、推理、反向链接等获得的运行时生成数据的临时数据层。此层还包含从文本节点推断出的语义知识。
Temporary Data Layer. A temporary data layer for external RDF data and runtime generated data gained from unstructured text knowledge, reasoning, inverted links, etc. This layer also contains semantic knowledge that is inferred from text nodes.
元数据层。应用程序可以使用此层添加有关添加到 PKG 的所有资源的元数据，包括作者、创建日期、访问权限或显示属性（如突出显示、对齐方式等）。
Metadata Layer. This layer can be used by applications to add metadata about all the resources added to the PKG, including author, creation date, access rights, or display properties like highlights, alignment, etc.
PKG 的核心存储在数据层和架构层中，而其他层可能被视为辅助层，用于支持应用程序的性能（即临时数据层）或功能（即元数据层）。因此，在下文中，我们将描述如何在数据层和架构层中对数据进行建模，并区分两种形式的内容：（半）结构化知识和非结构化知识。
The core of the PKG is stored in the data and schema layers, while the other layers might be considered as auxiliary to support aspects of the performance (i.e., the temporary data layer) or the functionality (i.e., the metadata layer) of the application. Therefore, in the following, we describe how data is modeled in the data and schema layers and distinguish between two forms of content: (semi-)structured knowledge and unstructured knowledge.
使用 SWS 建模（半）结构化 PKG 知识
Modeling (Semi-)Structured PKG Knowledge with SWSs
在我们的方法中，PKG 可以包含半结构化知识来描述实体，这些实体可以是个体、类、谓词等。这些描述通常被称为半结构化的，而不是结构化的，因为用于描述实体的属性和关系可能因知识元素而异，这意味着它们不遵循严格的模式。
In our approach, the PKG can contain semi-structured knowledge to describe entities, which could be an individual, a class, a predicate, etc. These descriptions are commonly referred to as semi-structured, and not structured, as the attributes and relations used to describe entities may vary from one knowledge element to another, meaning they don’t subscribe to a strict schema.
在 PKG 的背景下，半结构化知识允许在不对数据施加严格约束的情况下组织 PKG 中的概念，以及明确地互连不同的知识元素。此外，高级用户或为 SWS 开发的工具/插件可以使用 SPARQL 等语言查询这种半结构化知识，以帮助非专家用户轻松制定查询。
In the context of a PKG, semi-structured knowledge allows organizing concepts in the PKG without imposing rigorous constraints over the data, as well as interconnecting different knowledge elements explicitly. In addition, this semi-structured knowledge can be queried using a language like SPARQL either by advanced users or by tools/plug-ins developed for SWSs to assist non-expert users to easily formulate queries.
为了在我们的方法中表示半结构化知识，下面我们将介绍如何使用 SWS 对 PKG 中出现的知识元素进行建模，以及这些元素在我们提出的架构的哪一层中持久化。
To represent semi-structured knowledge in our approach, in the following we present how the knowledge elements that occur in a PKG are modeled with SWS and in which layer of our proposed architecture these elements are persisted.
表 8.2
Table 8.2
为了说明如何将 PKG 知识元素存储为 RDF，请考虑以下示例。在左侧，我们看到使用半结构化知识的 PKG 实体 Harry 的 UI（页面）。该实体通过 IRI http://example.org/Harry 进行标识。此实体是 Node 类的元素，对应于 RDF/S 中的 rdfs:Resource。它还属于 Person 和 Wizard 类，它们是用户定义的类。然后，我们有用于描述实体的属性。在这里，值的类型显示为文本，它在 RDF 中被转换为文字。最后，我们有建立与 RDF 图中其他节点的连接的关系。
To illustrate how the PKG knowledge elements are stored as RDF, consider the following example. To the left we see the UI (Page) for a PKG entity Harry using semi-structured knowledge. The entity is identified with the IRI http://example.org/Harry. This entity is an element of the class Node which corresponds to rdfs:Resource in RDF/S. It’s also of class Person, and Wizard, which are user-defined classes. Then, we have the attributes used to describe the entity. Here, the types of the values are shown as Text which is translated to a literal in RDF. Lastly, we have the relations that establish connections to other nodes in the RDF graph.
图 8.2
Figure 8.2
使用 SWS 表示 PKG 半结构化知识。左图：实体 Harry 的一页，其中包含半结构化知识，用于提供有关实体的描述。右图：描述实体 Harry 的底层 RDF 图的一部分。
Representing PKG semi-structured knowledge using SWS. Left: A page of the entity Harry with semi-structured knowledge to provide descriptions about the entity. Right: Part of the underlying RDF graph that describes the entity Harry.
使用 SWS 对非结构化 PKG 知识进行建模
Modeling Unstructured PKG Knowledge with SWS
一般而言，非结构化知识对应于文本、图像、音频或视频。通常，非结构化知识缺少任何类型的（机器可读）元数据或语义。在本节中，我们重点介绍底层 RDF 图中文本的表示，因为文本信息通常出现在 PKG 中有关实体的注释或描述中。
Generally speaking, unstructured knowledge corresponds to text, images, audio, or video. In general, unstructured Knowledge is missing any kind of (machine readable) metadata or semantics. In this section, we focus on the representation of text in the underlying RDF graph as textual information often appears in notes or descriptions about entities in the PKG.
在 RDF 图中建模文本
Modeling Text in an RDF Graph
文本，更具体地说是文本文档，可以看作是一棵树，其中的信息以分层方式存储。在这棵树中，文档对应于根，中间节点是将文本结构化为章节、节等的标题，最后内容（段落、图像等）是叶节点。可以用内容包含在层次结构各部分中的顺序来标记边缘。图 8.3 显示了这种基于树的文档表示的示例。
Text and, more concretely a text document, can be seen as a tree where information is stored in a hierarchical way. In this tree, the document corresponds to the root, the intermediate nodes are headings that structure the text in chapters, sections, etc., and finally the content (paragraphs, images, etc.) are leaf nodes. The edges can be labeled with the order in which the content is included in parts of the hierarchy. The figure 8.3 shows an example of this tree-based representation of a document.
图 8.3 将线性信息表示为文本（左）和有序树（右）
Figure 8.3 Representing linear information as text (left) and ordered tree (right)
在此表示中，中间节点和叶节点都表示为文本节点。在我们的模型中，这将作为 IRI 存储在 PKG 的底层 RDF 图中。这与多个 PKG 工具中对文本块的处理一致。除了文本节点之外，还必须考虑以下两个方面，以便从 RDF 三元组完全重构文本：
In this representation, both intermediate nodes and leaf nodes are denoted text nodes. In our model this would be stored in the underlying RDF graph of the PKG as IRIs. This is consistent with the handling of text blocks in several PKG tools. Besides the text nodes, it is important to consider the following two aspects to fully reconstruct the text from the RDF triples:
文本信息的层次结构（以标题、列表的形式）必须保留在 RDF 图中。
The hierarchy of the textual information (in the form of headings, listings) has to be preserved in the RDF graph.
文本出现的顺序必须在 RDF 图中编码。回想一下，RDF 图是一组三元组，即，三元组的序列化顺序并不代表三元组的实际显示顺序。因此，文本节点的顺序也必须以三元组的形式存储在 RDF 图中。
The order in which the text occurs has to be encoded in the RDF graph. Recall that an RDF graph is a set of triples, i.e., the order in which the triples are serialized does not represent an actual order how triples should be displayed. Therefore, the order of the text nodes has to be stored in the RDF graph as well in the form of triples.
要对 RDF 图中的信息层次和顺序进行编码，有几种方法，在表示的详细程度和插入或删除文本节点等操作的效率方面各有不同。为了实现它们，可以将 RDF/S 定义与都柏林核心词汇表或 Hydra 词汇表结合使用。
To encode the hierarchy and order of information in an RDF graph, there are several ways with different tradeoffs in terms of the verbosity of the representation and the efficiency of operations like insertion or deletion of text nodes. For implementing them, one could use RDF/S definitions in combination with a Dublin Core vocabulary, or the Hydra vocabulary.
进一步的实施选项。除了 RDF/S、Dublin Core 和 Hydra 之外，还有其他词汇表，它们要么兼容，要么直接建立在语义网标准之上，也可用于表示文本节点。下面，我们列出了其中一些词汇表。
Further implementation options. In addition to RDF/S, Dublin Core, and Hydra, there are other vocabularies that are either compatible or directly built on top of Semantic Web standards that can also be used to represent text nodes. In the following, we list some of these vocabularies.
关联数据平台 (LDP) 为更新 RDF 资源提供了基础。具体来说，它将 LDP 容器的概念定义为 RDF 文档的集合。此定义还可用于表示 PKG 中的文本节点。
The Linked Data Platform (LDP) provides the foundations to update RDF resources. In particular, it defines the notion of LDP containers as a collection of RDF documents. This definition could also be used to represent text nodes in a PKG.
TREE 超媒体规范提供了对分层信息进行建模的定义，并且一些提出的概念也与 Hydra 和 LDP 兼容。
The TREE hypermedia specification provides definitions to model hierarchical information and some of the proposed concepts are also compatible with Hydra and LDP.
语义化 Markdown 扩展
A Semantic Markdown Extension
在上一节中，我们介绍了在 RDF 图中表示文本文档的方法，以便存储非结构化文本知识（注释、评论等）。从这个意义上讲，我们专注于捕捉文本的层次结构和顺序，并假设无法从文本中提取更多信息。但是，大多数 PKG 工具都使用 Markdown 为文本添加某些格式。Markdown 是一种用于创建格式化文本的轻量级标记语言。使用 Markdown，纯文本可以通过在文本前面添加或括在特殊字符（如 \\ # * - 等）中来转换。虽然 Markdown 尚未标准化，但它受到网络社区的欢迎并不断扩展。在本节中，我们提出了一种 Markdown 的语义扩展，除了格式化方面之外，它还允许纯文本嵌入语义信息，这些信息稍后可以转换为 RDF 三元组。为了解释我们的扩展，我们首先回顾了几种 Markdown 风格——基本、扩展和“超文本”，然后解释如何使用简单的注释，PKG 用户也可以在文本描述中包含语义数据。
In the previous section we presented our approach to represent text documents in an RDF graph, to enable storage of unstructured text knowledge (notes, comments, etc.). In this sense, we focus on capturing the aspects of hierarchy and ordering of text, and assume that no further information could be extracted from the text. However, most PKG tools use Markdown to add certain formatting to the text. Markdown is a lightweight markup language for creating formatted text. With Markdown, plain text gets transformed by prepending or enclosing it in special characters like \\ # * - etc. Although not standardized, Markdown is embraced by the web community and continuously extended. In this section, we propose a semantic extension for Markdown that in addition to formatting aspects, allows plain text to embed semantic information that can later be translated into RDF triples. To explain our extension, first we revisit several Markdown flavors – basic, extended, and “hypertext” – and then explain how, using simple annotations, users of PKG can also include semantic data within textual descriptions.
基本 Markdown。主要包括文本格式：
Basic Markdown. Includes mostly text formatting:
# 标题，**粗体**，*斜体*，~~删除线~~，`代码`
# Headings, **Bold**, *Italic*, ~~Strikethrough~~, `Code`
编号和项目符号列表
Numbered and bulleted lists
图像和链接作为 URL 引用
Images and links as URL references
扩展 Markdown。包括高级格式选项：
Extended Markdown. Includes advanced formatting options:
表格
Tables
标题 ID（用于文档内导航）
Heading IDs (for in-document navigation)
语法高亮的代码块
Syntax highlighted code blocks
脚注
Footnotes
待办事项
To-dos
表情符号
Emoji
突出显示
Highlighting
下标和上标
Sub- and superscript
目录
Table of content
标注
Callouts
评论
Comments
字幕
Captions
“超文本 Markdown”。最近，一些笔记工具扩展了 Markdown 语法，启用了全工具超链接 [Bear、Notion、Roam、LogSeq、Obsidian]。用某些特殊字符（@、[[]]、{{}} 等）括起来的文本会自动被索引为知识元素之间的特殊连接。这些将呈现为链接、提及或嵌入。
“Hypertext Markdown”. Recently, some note-taking tools have extended the Markdown syntax, enabling toolwide Hyperlinks [Bear, Notion, Roam, LogSeq, Obsidian]. Text enclosed in certain special characters (@, [[]], {{}}, etc.) is automatically indexed as special connections between knowledge elements. These are rendered as links, mentions or embeds.
[[ 转换为文档链接
[[ gets converted to links to documents
（（链接到一个段落
(( links to a paragraph
{{ 嵌入文档或段落
{{ embeds documents or paragraphs
$$ LaTeX 代码
$$ LaTeX code
^^ 突出显示文本
^^ Highlight text
语义 Markdown。我们提出的解决方案采用超文本 Markdown 的方法，并向其中添加语义关系。然后，这些语义关系被转换为 RDF 三元组，并在运行时体现到临时数据层中。通过将表达式包装在特殊字符中来调用转换为 RDF。这些字符被输入两次，因此不会在应用程序中意外触发。我们得出了以下示例列表：
Semantic Markdown. Our proposed solution takes the approach of Hypertext Markdown and adds semantic relationships to it. These semantic relationships are then converted into RDF triples and manifested into the temporary Data Layer during runtime. The conversion to RDF is invoked by wrapping the expressions in special characters. These characters are typed twice, so they do not get accidentally triggered in the application. We came up with the following exemplary list:
[[label/IRI]] 创建指向具有此标签或 IRI (rdfs:seeAlso) 的实体的有向链接。
[[label/IRI]] create a directed link to an entity with this label or IRI (rdfs:seeAlso).
((label/IRI)) 嵌入实体或语义 markdown 包括嵌套 markdown。
((label/IRI)) embed an entity or semantic markdown including nested markdown.
{{predicate object}} 将此谓词和对象附加到语义 markdown 的父实体。
{{predicate object}} attach this predicate and object to the parent entity of the semantic markdown.
<<IRI>> 链接到外部 RDF 实体
<<IRI>> link to external RDF entity
<<IRI，谓词>>链接到外部 RDF 实体属性。
<<IRI, predicate>> link to external RDF entity attribute.
数据层的 RDF 三元组与语义 Markdown 中的 RDF 语句之间的区别有两个方面：事实层的 RDF 三元组表示机器可读且可查询的示意性半结构化事实知识。语义 Markdown 位于人类可读的文本中，并且经常会发生变化和重构。它可以随时保存到事实层中。
The difference between RDF triples from the data layer and RDF statements in the semantic markdown, is twofold: RDF triples in the facts layer represent schematic semi-structured factual knowledge that is machine readable and can be queried. Semantic Markdown is inside human readable text and subject to frequent changes and refactoring. It can at any point be persisted into the facts layer.
{{ rdf:type rdfs:Resource; }}
{{ rdf:type rdfs:Resource; }}
（使用语义 Markdown 扩展的文本示例）
(Example of a text using the semantic Markdown extension)
为了说明 RDF 三元组如何嵌入 {{}}，让我们考虑以下示例：
To illustrate how RDF triples are embedded with {{}}, let us consider the following example:
# 关于实体的注释：地球
# Notes on the entity :Earth
地球是距离 {{ rdfs:seeAlso :Sun }} 最近的 {{ rdf:type :Planet }}，并且
Earth is the third {{ rdf:type :Planet }} from the {{ rdfs:seeAlso :Sun }} and
唯一已知存在生命的天文物体。
the only astronomical object known to harbor life.
尽管在整个 {{ ex:location ex:Solar_System }} 中都能发现大量的水，但只有地球才维持着液态表面[[水]]。
While large volumes of water can be found throughout the {{ ex:location ex:Solar_System }}, only Earth sustains liquid surface [[water]].
使用语义 Markdown 扩展的文本示例
Example of a text using the semantic Markdown extension
此语义标记将在临时数据层中创建以下三元组，
This semantic markdown would create the following triples in the temporary data layer,
:地球 rdf:类型 :行星。
:Earth rdf:type :Planet .
:地球 rdfs:seeAlso :太阳 。
:Earth rdfs:seeAlso :Sun .
:地球例如:位置:太阳系。
:Earth ex:location :Solar_System .
从语义 Markdown 文本中提取 RDF 三元组并存储在数据层中
RDF triples extracted from the semantic Markdown text and stored in the Data Layer
并在 UI 中呈现如下形式：
And would be rendered in the ui like this:
地球是距太阳第三远的行星，
Earth is the third planet from the sun and
唯一已知存在生命的天文物体。
the only astronomical object known to harbor life.
虽然整个太阳系都拥有大量的水，但只有地球才拥有液态地表水。
While large volumes of water can be found throughout the solar system, only Earth sustains liquid surface water.
UI 中的显示示例
Example of display in the UI
链接和 rdf 三元组可以内联使用，而嵌入只能用作独立的语义 markdown 节点。
Links and rdf triples can be used inline, while embeds can only be used as standalone semantic markdown nodes.
处理基于 RDF 的 PKG：CRUD 操作
Handling RDF-based PKGs: CRUD Operations
在本节中，我们概述了创建、读取、更新和删除 (CRUD) 操作对所提数据模型的不同元素和层的影响。为此，我们使用 SPARQL 查询语言来展示如何在使用 RDF 建模的 PKG 上实现这些操作。
In this section, we outline the effects of create, read, update and delete (CRUD) operations on the different elements and layers of the proposed data model. For this, we use the SPARQL query language to show how the operations are implemented over the PKG modelled with RDF.
创建。在我们提出的模型中，创建操作允许向 PKG 添加不同类型的元素，即类、关系或“非本体”节点。为此，每个节点都分配给通用类 rdfs:Resource，并始终使用 rdfs:label 用人类可读的名称进行注释。对于类和属性，它们还用各自的类 rdfs:Class 或 rdf:Property 进行注释。在下文中，我们将展示如何使用 SPARQL 实现这些操作。
Create. In our proposed model, the operation of create allows for adding different types of elements to the PKG, i.e., classes, relationships, or “non-ontological” nodes. For this, every node is assigned to the general class rdfs:Resource and always annotated with a human-readable name using rdfs:label. In the case of classes and properties, these are additionally annotated with their respective classes rdfs:Class or rdf:Property. In the following, we show how these operations are implemented using SPARQL.
插入数据
INSERT DATA
{ 图：用户架构 {
{ GRAPH :UserSchema {
:C rdf:类型 rdfs:资源。
:C rdf:type rdfs:Resource .
:C rdf:类型 rdfs:类。
:C rdf:type rdfs:Class .
：C rdfs：标签“想法”。
:C rdfs:label "Idea" .
} }
} }
插入数据
INSERT DATA
{ 图：用户架构 {
{ GRAPH :UserSchema {
:R rdf:类型 rdfs:资源。
:R rdf:type rdfs:Resource .
:R rdf:类型 rdf:属性。
:R rdf:type rdf:Property .
：R rdfs：标签“作者”。
:R rdfs:label "author" .
} }
} }
插入数据
INSERT DATA
{ 图表：数据 {
{ GRAPH :Data {
:n rdf:类型 rdfs:资源。
:n rdf:type rdfs:Resource .
:n rdfs:label “假设”。
:n rdfs:label "Hypothesis" .
} }
} }
插入数据
INSERT DATA
{ 图表:数据 { :n :R *val* . } }
{ GRAPH :Data { :n :R *val* . } }
读取。读取操作对应于检索与节点相关联的信息的 SPARQL SELECT 查询。节点可以位于 PKG 的数据层中，也可以位于架构层中。我们区分两种类型的读取操作。第一种只是检索具有给定标题的节点。第二种是检索节点出现的所有语句。
Read. The operation of read corresponds to SPARQL SELECT queries that retrieve information associated with a node. The node can be either in the data or in the schema layer of the PKG. We distinguish two types of read operations. The first one is just about retrieving nodes with a given title. The second is about retrieving all the statements where the node occurs.
选择 *
SELECT *
来源：数据
FROM :Data
来自：UserSchema
FROM :UserSchema
在哪里 {
WHERE {
?s rdfs:label “假设”。
?s rdfs:label "Hypothesis" .
}
}
选择 *
SELECT *
来源：数据
FROM :Data
来自：UserSchema
FROM :UserSchema
在哪里 {
WHERE {
?x rdfs:label “假设”。
?x rdfs:label "Hypothesis" .
{ ?x ?p ?o . }
{ ?x ?p ?o . }
联盟
UNION
{ ?s ?p ? x. }
{ ?s ?p ? x. }
}
}
更新。PKG 上的更新在 SPARQL 中实现为 DELETE 和 INSERT 操作序列。在我们的模型中，更新可以指 RDF 三元组对象的更改，也可以指重命名数据层或架构层中节点的标题。以下使用 SPARQL 显示了这些操作。
Update. Updates over the PKG are implemented in SPARQL as a sequence of DELETE and INSERT operations. In our model, updates can refer to changes in the object of an RDF triple or to renaming the title of a node in the data layer or in the schema layer. These operations are shown in the following using SPARQL.
数据
WITH :Data
删除 {:n:R *val*.}
DELETE { :n :R *val* . }
插入 {:n:R *newval*.}
INSERT { :n :R *newval* . }
数据
WITH :Data
删除 {:n rdfs:label "title*"* . }
DELETE { :n rdfs:label "title*"* . }
插入{:n rdfs:label "新标题*"*.}
INSERT { :n rdfs:label "new title*"* .}
使用 :UserSchema
WITH :UserSchema
删除 {:n rdfs:label "title*"* . }
DELETE { :n rdfs:label "title*"* . }
插入{:n rdfs:label "新标题*"*.}
INSERT { :n rdfs:label "new title*"* .}
删除。删除操作可以在数据层特定节点的语句级别执行，也可以在 PKG 的词汇表/本体级别执行。下面，我们将说明如何在模型中定义删除以及如何使用 SPARQL 实现删除，以及相应地从数据层或架构层删除哪些元素。
Delete. The deletion operation can be performed at the level of statements in specific nodes of the data layer, or at the level of the vocabulary/ontology of the PKG. In the following, we specify how deletions are defined in our model and implemented using SPARQL and what elements are deleted from the data layer or the schema layer accordingly.
删除数据 { 图表:数据 { :n :R *val* .}}
DELETE DATA { GRAPH :Data { :n :R *val* .}}
数据
WITH :Data
删除 {:n?p?o.}
DELETE { :n ?p ?o . }
其中 {:n?p?o.}
WHERE { :n ?p ?o . }
数据
WITH :Data
删除 { ?s ?p :n . }
DELETE { ?s ?p :n . }
其中 { ?s ?p :n . }
WHERE { ?s ?p :n . }
数据
WITH :Data
删除 { ?s :R ?o . }
DELETE { ?s :R ?o . }
其中 { ?s :R ?o . }
WHERE { ?s :R ?o . }
使用 :UserSchema
WITH :UserSchema
删除 { ?x ?p ?o . }
DELETE { ?x ?p ?o . }
其中 { ?x rdf:type rdf:Property . 过滤器 (?x = :R)}
WHERE { ?x rdf:type rdf:Property . FILTER (?x = :R)}
使用 :UserSchema
WITH :UserSchema
删除 { ?s ?p ?x . }
DELETE { ?s ?p ?x . }
其中 { ?x rdf:type rdf:Property . 过滤器 (?x = :R)}
WHERE { ?x rdf:type rdf:Property . FILTER (?x = :R)}
数据
WITH :Data
删除 { ?s rdf:type :C . }
DELETE { ?s rdf:type :C . }
其中 {?s rdf:type:C.}
WHERE { ?s rdf:type :C .}
使用 :UserSchema
WITH :UserSchema
删除 { ?x ?p ?o . }
DELETE { ?x ?p ?o . }
其中 { ?x rdf:type rdfs:Class .过滤器 (?x = :C)}
WHERE { ?x rdf:type rdfs:Class . FILTER (?x = :C)}
使用 :UserSchema
WITH :UserSchema
删除 { ?s ?p ?x .}
DELETE { ?s ?p ?x .}
其中 { ?x rdf:type rdfs:Class .过滤器 (?x = :C)}
WHERE { ?x rdf:type rdfs:Class . FILTER (?x = :C)}
表示层
Presentation Layer
在本节中，我们将讨论如何将模型应用于用户界面 (UI)。我们将使用一个原型实现来演示我们的方法，该实现可在 https://standardpkgs.github.io/arc/ 上访问。
In this section we will discuss applying the model onto a user interface (UI). We will demonstrate our approach with a prototypical implementation that can be accessed at https://standardpkgs.github.io/arc/.
此原型仅用于提供可能的视觉表示示例和演示基本导航和编辑。它并非成品或可用软件。
This prototype is just provided to give an example of a possible visual representation and demo basic navigation and editing. It does not intend to be finished or usable software.
我们将解释如何将我们的模型映射到 UI 上，并展示具有此 UI 的软件的原型实现。
We will explain how our model can be mapped onto a UI and show a prototypical implementation of a software with this UI.
显示模型
Displaying the model
RDF 资源可分为实体、类和关系。实体代表各自域中的个体，而类则类似于实体集。关系是资源之间的有向链接。在我们的原型中，实体具有其他两个继承的基本 UI，其中包括呈现实体的 rdfs:label、IRI、类和关系。类还将显示其唯一元素（以此类作为其 rdf:type 的资源）。关系还显示它们所属的完整三元组，以及作为关系三元组中的主体或客体出现的唯一资源。
RDF Resources can be categorized as Entities, Classes and Relationships. Entities represent individuals in the respective Domain, while Classes are like sets of entities. Relationships are directed links between resources. In our Prototype, entities feature the bare-bone UI that the other two inherit, which includes rendering the rdfs:label, IRI, Classes and Relationships of the entity. Classes will also display their unique elements (resources that have this class as their rdf:type). Relationships additionally display the full triples they are part of, and the unique resources that appear as subjects or objects in triples with the relationship.
可视化实体
Visualizing entities
对于实体 IRI，首先呈现的是人类可读的 rdfs:label，通过查找三元组“IRI rdfs:label <Label>”。如果未找到标签三元组，则显示 IRI。在其下方，IRI 的类别被呈现为药丸（类似于其他 UI 中类别或标签的显示），通过查询所有形式为“IRI rdf:type <Class>”的三元组。接下来，所有以 IRI 为主题、以文字为对象的三元组都呈现在名为“属性”的部分（owl:datatypeproperty）中。在其下方，所有以当前 IRI 为主题、以节点为对象的三元组都显示在“关系”下。在页面底部，所有语义标记都呈现为交互式大纲 UI。
For an entity IRI, the first thing rendered is the human readable rdfs:label, by looking up the triple “IRI rdfs:label <Label>”. If no label triple is found, the IRI is displayed. Below that the classes of the IRI are rendered as pills (similar to displays of categories or tags in other UIs), by querying for all triples of the form “IRI rdf:type <Class>”. Next all the triples with the IRI as subject and literals as objects are rendered in a section called “Attributes” (owl:datatypeproperty). Below that all triples with the current IRI as subject and Nodes as objects are displayed under “Relationships”. At the bottom of the page all semantic markdown is rendered as an interactive outliner UI.
主编辑器部分的旁边是显示传入关系和对 IRI 的提及的旁注部分。这些是查询“<all> <all> CRI”形式的三元组的结果。
To the side of the main Editor section is an aside section that displays incoming relationships and mentions to the IRI. These are the results of querying for triples of the form “<all> <all> CRI”.
图 8.4 实体编辑器（通用 UI）
Figure 8.4 Entity Editor (general UI)
可视化类别
Visualizing classes
类具有实体所具有的所有部分，此外还在“实例”下呈现其唯一元素（“<Node> rdf:type IRI”形式的三元组）
Classes have all the sections that entities have and in addition render their unique elements (triples of the form “<Node> rdf:type IRI”,) under “Instances”
图 8.5 类编辑器
Figure 8.5 Class Editor
可视化关系
Visualizing relationships
关系通过在“三元组”下显示谓词位置上的所有三元组来扩展实体 UI。此外，它们还通过在“主体”和“客体”下查找“<Node> IRI ”和“ IRI <Node>”形式的三元组来显示 2 个部分，这些部分具有唯一的主体和客体的节点。
Relationships extend the entity UI by displaying all triples with them in the predicate position under “triples”. In addition to that they display 2 sections with nodes that are unique subjects and objects, by looking for triples of the form “<Node> IRI ” and “ IRI <Node>” under “subjects” and “objects”.
图 8.6 属性编辑器
Figure 8.6 Property Editor
在研究了 DBpedia 和当前 PKG 工具如何呈现其 UI 之后，我们得出了这种表示，同时也考虑到使用这种 UI 的人不一定对 SWS 有任何先验知识。
We arrived at this representation after looking at how DBpedia and current PKG Tools render their UI, while also taking into consideration that the people who would use such a UI don’t necessarily have any prior knowledge about SWS.
以这种方式呈现 RDF 数据不是强制性的，还有许多其他方式可以选择显示有关 RDF 资源的数据。举一些例子，请考虑现有的可视化，如 wikidata、DBpedia、DBpedia LODvisualizer、Obsidian 图形可视化等。
Rendering the RDF Data in this way is not mandatory and there are many other ways one could choose to display data about RDF resources. To give some examples consider existing visualizations like wikidata, DBpedia, DBpedia LODvisualizer, Obsidian graph visualization, etc.
可用性考虑
Usability Considerations
虽然 PKG 的许多挑战可以通过数据模型和架构来解决，但 SWS 技术的用户体验 (UX) 问题需要单独考虑。即使对于计算机科学家来说，语义网技术也很难理解 [easierRDF]。期望非技术背景的人掌握 SWS 所依赖的所有概念是不现实的。因此，对于个人使用，需要用不言自明的 UI 抽象出这些概念。在本节中，我们介绍了一个实现我们模型的原型，旨在不依赖于先前的语义网技术知识。
While many challenges of PKGs can be addressed with the data model and architecture, SWS technologies' User Experience (UX) problems need to be considered separately. Semantic web technologies are hard to understand even for Computer Scientists [easierRDF]. Expecting people from a non-technical background to grasp all the concepts that SWSs rely on is unrealistic. Thus, for personal use these concepts need to be abstracted away with a self-explanatory UI. In this section, we present a prototype that implements our model and aims to not rely on previous knowledge of semantic web technologies.
该原型以单页反应式 Web 应用程序的形式实现。这样无需下载或配置客户端即可实现无摩擦设置。我们的目标是，即使该应用程序基于我们的 RDF 数据模型和 SWS，也无需任何语义 Web 技术知识即可使用该应用程序。
The prototype is implemented as a single-page reactive web application. This enables a frictionless setup without the need to download or configure a client. The goal is for no knowledge of semantic web technologies to be necessary to use the app, even though it is based on our RDF data model and SWSs.
用户界面
The UI
图 8.7 原型的 UI
Figure 8.7 The UI of the prototype
上面的图 8.7 显示了 UI 的屏幕截图。它分为几个区域，其中一些类似于人们经常使用的应用程序中的部分，例如笔记应用程序、维基百科或电子邮件客户端。从左上角到右下角：
Figure 8.7 above shows a screenshot of the UI. It is divided into several areas, some of them similar to sections found in apps people frequently use, like note-taking apps, Wikipedia or email clients. From top left to bottom right:
工具栏。工具栏与现代浏览器中的工具栏类似。它在选项卡中显示打开的节点，IRI 显示为选项卡的标题。最外角是用于访问重要功能的快捷方式。
Toolbar. The toolbar is similar to the ones found in modern browsers. It displays open nodes in tabs with the IRI appearing as the title for the tab. In the outermost corners are shortcuts for accessing prominent functionality.
侧边栏。侧边栏用作导航和定位。用户可以导航到 PKG 中的类、关系和他们最喜欢的节点。还有已使用架构和集/视图的概述。
Sidebar. The sidebar serves as navigation and orientation. Users can navigate to classes, relationships and their favorite nodes in the PKG. There are also overviews of used schema and sets / views.
编辑器。编辑器呈现当前活动节点的所有相关信息。用户可以切换与自己相关的信息并导航到相关节点。他还可以创建、更新和删除相关实体和文本。
Editor. The editor renders all the relevant information about the currently active node. The user can toggle which information is relevant to him and navigate to related nodes. He can also create, update and delete related entities and text.
图形面板。图形面板以可视化形式概述当前和相关节点的上下文。这为 PKG 提供了进一步的方向和空间连通感。
Graph panel. The graph panel serves as an overview of the context of the current and associated nodes in visualized form. This provides further orientation and a spatial sense of connectedness in the PKG.
关系面板。关系面板显示推断的关系和提及。（意思是从其他节点到当前节点的弧。当前节点的标签和类被放置在特殊位置，并从关系中隐藏。标签以标题的形式突出显示，让人想起大多数文本处理器。类显示在它正下方，类似于其他应用程序中的标签或类别。这样做是为了建立对用户可能与之交互的类似 UI 的熟悉度。
Relation panel. The relation panel shows inferred relations and mentions. (Meaning arcs from other nodes to the current node. The label and classes of the current node are placed in special positions and hidden from the relations. The label is prominently displayed as the title, reminiscent of most text processors. Classes are presented right below it, similar to tags or categories in other applications. This is done to build on the familiarity with similar UIs the User might have interacted with.
请注意，类、属性、实体和文字在 UI 中是如何在视觉上区分的，这突出显示了它们的不同性质，并提供了轻松的内容导航。语义标记显示在大纲 UI 中，就像它是当前节点的分层附件一样。其中的结构化语义以不透明的方式显示在相关节点上，以突出显示它们不是在数据层中体现的，而是从注释中推断出来的。按钮使用户能够将它们转换为从注释层到数据层的显式语句。集成外部结构化数据也具有类似的功能。
Notice how classes, properties, entities and literals are all visually distinguished in the UI, highlighting their different nature and providing easy navigation of the content. The semantic markdown is displayed in an outliner UI, as though it was a hierarchical attachment to the current node. The structured semantics in it are displayed on the relevant nodes in an opaque fashion, to highlight the fact that they are not manifested in the data layer, but rather inferred from notes. A button enables the user to transform them into explicit statements from the note layer into the data layer. Similar functionality is present for integrating external structured data.
功能
Functionality
以下是原型的特性和功能概述。所有更改均保存在内存中的 RDF 四元组存储中。
The following is an overview of features and functionalities of the prototype. All of the changes are persisted into the in-memory RDF quad store.
创建节点。按下左下角标有“+ 新节点”的按钮或使用键盘快捷键“ctrl + n”可在图形中创建新节点。用户会自动导航到新节点并可以编辑标题/标签。
Create Node. Pressing the button labeled “+ New Node” in the bottom left corner or using the keyboard shortcut “ctrl + n” creates a new node in the graph. The user is automatically navigated to the new node and can edit the title / label.
附加类。可以通过单击标题/标签正下方的“+ 添加类”按钮将新类附加到当前节点。
Attach Class. A new class can be attached to the current node by clicking the “+ add Class” button right below the title / label.
附加属性/关系。通过按相应部分底部的“+ 添加属性/关系”按钮，可以创建当前节点处于主题位置的 RDF 三元组。这会提示用户输入三元组的谓词（关系）和对象（目标实体/文字）的标签。
Attach Attribute / Relationship. An RDF triple with the current node in subject position can be created by pressing the “+ add attribute / relation” button at the bottom of the respective sections. This prompts the user to enter a label for the predicate (Relationship) and Object (target entity / literal) of the triple.
编写语义化 Markdown。在注释部分，可以在大纲文本处理器界面中编辑语义化 Markdown（此功能尚未完全实现）。
Write Semantic Markdown. In the notes section, semantic markdown can be edited in an outliner text processor interface (This feature is not fully implemented yet).
切换 RDF 图形数据集。用户可以在侧边栏中标有选择输入的“图形”中选择 RDF 数据集，以在图形数据集之间切换。
Switch RDF Graph Dataset. The user can select the RDF dataset in the “Graph” labeled select input in the Sidebar, to switch between graph datasets.
从 URL 导入 RDF 图表。用户可以从外部源导入 RDF 数据，方法是从侧边栏中标有选择输入的“图表”中选择“+ 从 URL 导入”。
Import RDF Graphs. from URL The user can import RDF data from external sources, by selecting “+ import from URL” from the “Graph” labeled select input in the sidebar.
结论与展望
Conclusions and Outlook
在本章中，我们研究了使用基于标准的 PKG 来解决信息过载问题的个人知识管理工具的基础。我们确定基于专有格式或文本文件的 PKM 工具不适合长期个人知识管理。
In this chapter, we have investigated the foundations for personal knowledge management tools using standards based PKGs to combat the information overload problem. We established that PKM tools based on proprietary formats or text files are not viable candidates for long term personal knowledge management.
我们提出了使用 RDF 作为 PKG 的数据模型，并概述了如何使用它存储结构化和非结构化数据。为了实现这一点，我们提出了开发一个可以表达 RDF 语句的 markdown 扩展。作为概念验证，我们实现了一个 Web 应用程序原型，它可以将结构化和非结构化知识存储为 RDF 图。
We have proposed RDF as the data model for PKGs and outlined how to store both structured and unstructured data with it. To achieve this, we proposed the development of a markdown extension that can express RDF statements. As a proof of concept, we implemented a web application prototype that can store structured and unstructured knowledge as an RDF graph.
原型作为模型和表示层的概念证明，基于内存中的数据集。这显然无法扩展。虽然有多个（图形）数据库可以作为此类应用程序的“后端”，但在这里我们想重点介绍一个专注于开放标准化数据格式的项目，同时解决隐私和数据所有权问题，我们认为这非常重要。
The Prototype serving as a proof of concept of the model and presentation layer, was based on a dataset in-memory. This obviously won’t scale. While there are several (Graph) Databases that could serve as a “backend” for an application like this, here we would like to highlight a project which focuses on open standardized data formats while addressing privacy and data ownership concerns, which we find quite important.
SOLID 项目是 Tim Berners Lee 发起的一项计划，旨在为个人提供在线数据存储，即所谓的“Pod”。这些 pod 可能作为实现 PKG 生态系统愿景的后端。该项目已经支持 RDF 数据模型，并且与多种图形查询语言兼容。
The SOLID project is an initiative by Tim Berners Lee that aims to provide individuals with online data stores, so-called “Pods.” These pods could potentially serve as a backend for enabling the vision of a PKG ecosystem. The project already supports the RDF data model, and is compatible with several graph query languages.
笔记
Notes
[1]万维网联盟https://www.w3.org/
[1] World Wide Web Consortium https://www.w3.org/
[2] 语义网。https://www.w3.org/standards/semanticweb/
[2] Semantic Web. https://www.w3.org/standards/semanticweb/
[3] https://docs.google.com/document/d/1CEvxOoYMh9VvZJOikbjODQccA4tbg0YwjZC-3bN9I3o/edit?usp=sharing
[3] https://docs.google.com/document/d/1CEvxOoYMh9VvZJOikbjODQccA4tbg0YwjZC-3bN9I3o/edit?usp=sharing
第九章
Chapter 9
图像、个人知识和多模态图
Images, Personal Knowledge and Multi-Modal Graphs
玛格丽特·沃伦
介绍
Introduction
2018 年夏天，我去巴黎卢浮宫看《蒙娜丽莎》。我和其他几百人一起耐心等待，我们慢慢涌向绳索屏障，恭敬地等待着与达芬奇神秘微笑女人合影。
In the summer of 2018, I visited the Louvre in Paris to see the Mona Lisa. I patiently waited with several hundred other people as we all surged slowly toward the rope barrier, dutifully waiting to make my own selfie with DaVinci’s mysterious smiling woman.
当我站在绳子旁时，我无意中听到一对夫妻之间最精彩的对话。当丈夫给妻子拍照时，妻子说：“好吧，现在轮到你了。”丈夫回答说：“为什么？”妻子说：“因为你必须这么做！”他说：“为什么？蒙娜丽莎的照片有很多，你现在有一张。我知道我来过这里——我不需要照片来证明这一点。”接着她坚持要和他一起拍一张照片，我一边走一边咯咯笑着，她坚持要拍。我不知道他是否让步了，但很明显，她强烈要求丈夫必须把这一刻留在他生命的时间线上。
While I was at the rope, I overheard the most fabulous conversation between a couple. As the husband is taking a photo of his wife, she said, “Ok, now it’s your turn.” To which he replied, “Why?” and she said, “Because you have to!” and he said, “What for? There are plenty of photos of the Mona Lisa, you have one now. I know I was here – I don’t need a photo to prove it,” followed by her insistence that a photo be taken with him in it, and this continued as I moved along giggling to myself. I don’t know for sure whether he ever relented, but it was clear she had a strong urge that her husband must preserve this moment in his life’s timeline.
她并不是唯一一个有这种渴望的人。据一位讲解员说，每天有超过 15,000 张蒙娜丽莎的照片被上传到网上。
She is not alone in this urge. According to a docent, over 15,000 images of the Mona Lisa are uploaded to the web every single day.
在数码摄影和云存储出现之前，这类图像可能被存储在相册、鞋盒中，或者在可怕的家庭度假幻灯片中展示，然后被存放在阁楼的盒子里。
Prior to digital photography and cloud-based storage, images like this might have been stored in photo albums, shoe boxes or shown in the dreaded family vacation slideshow before being stored in a box in the attic.
但五年后的今天，我打开了 Google Photo 应用，搜索“蒙娜丽莎”文本字符串，找到了蒙娜丽莎的照片。虽然它没有检索到我的自拍照，但蒙娜丽莎太出名了，它检索到了我当天拍摄的其他照片。我很快就找到了我 2018 年 7 月 26 日参观过的卢浮宫。我还能立即访问那一周我在巴黎拍摄的所有其他体验。
But today, five years later, I was able to open my Google Photo app, search for the text string “Mona Lisa.” and find photos of it. While it didn’t retrieve my selfie, the Mona Lisa is so well known that it retrieved my other photos of it that day. I was quickly able to find I had visited the Louvre on July 26, 2018. I was also instantly able to access all the other experiences I had photographed in Paris that week.
如今，智能手机上的图像可以作为助记工具不断使用——神奇的回放功能可以让你回忆起很多重要时刻，无论大小。很明显，我们现在以越来越复杂的方式使用数字图像来增强我们的认知过程。
Our smartphone images are now continuously available as mnemonic aids – magical rewind features for many moments, big and small. It’s also clear that we now use our digital images in increasingly sophisticated ways to augment our cognitive processes.
我们拍摄收据、食谱和文件。我们拍摄想要买卖的产品或零件。我们拍摄收藏品和资产序列号。我们甚至可以扫描书架的照片，并使用键入的标题查找书籍。
We grab shots of receipts, recipes, and documents. We take photos of products or parts we want to buy or sell. We take photos of our collectibles and the serial numbers of assets we own. We can even scan photos of our bookshelves and find our books using a typed title.
这些用例中有许多是个人知识管理 (PKM) 的示例。根据维基百科，[1] 个人知识管理是一个收集信息的过程，人们在日常活动中使用它来收集、分类、存储、搜索、检索和共享知识。
Many of these use cases are examples of Personal Knowledge Management (PKM). According to Wikipedia, [1] personal knowledge management is a process of collecting information that a person uses to gather, classify, store, search, retrieve and share knowledge in their daily activities.
其他可以使用图像的 PKM 类型包括记笔记、存档个人历史信息、家谱记录和实物收藏品的管理，例如：黑胶唱片、古董车、模型火车、明信片、邮票、吉他、书籍、玩偶、钟表、手表——不胜枚举。过去，用户可能已将这些用途的数据记录在面向文本的数据库系统中。如今，他们可以轻松地拍摄这些信息，并使用基本的 AI 标记技术进行搜索和检索。
Other types of PKM that can use images can include note taking, archiving personal historical information, genealogical records and the curation of physical collections, such as: vinyl records, vintage cars, model trains, postcards, stamps, guitars, books, dolls, clocks, watches – the lists are endless. In the past, users might have recorded the data for these uses in text-oriented database systems. Today, they can easily photograph this information and use basic AI tagging techniques for search and retrieval.
然而，尽管计算机视觉取得了巨大进步，机器对许多类型图像的内容和背景的理解仍然不足。人工智能 (AI) 字幕软件可以识别火箭升空，但不知道这是阿波罗 11 号任务。另一方面，人类通常可以从图像中包含的元数据或图像周围的其他线索推断出这些信息。
But despite huge advances in computer vision, machine understanding of the content and context of many types of images is still lacking. Artificial Intelligence (AI) captioning software can recognize a rocket lifting into space, but not know that it was the Apollo 11 mission. A human, on the other hand, can often infer this information from metadata included with the image or other clues surrounding the image.
个人知识图谱 (PKG) 提供了一种以图形形式存储、分类、排序和检索个人知识本身的方法。PKG 技术背后的目标通常包括丰富知识图谱中存储的概念的语义。图像及其描述性内容对 PKG 非常有用，但经验表明，它们通常仅将图像用作图形中的插图。
Personal Knowledge Graphs (PKGs) provide a means to store, classify, sort and retrieve personal knowledge itself in a graph form. The objectives behind PKG technologies often include the semantic enrichment of the concepts stored in the knowledge graph. Images and their descriptive content could be immensely useful for PKGs, yet experience has shown that they often only use images as illustrations in the graph.
本章将探讨图像在 PKG 中扮演的角色的多个方面——无论是作为 PKG 设计的核心元素还是作为图形的辅助元素。它还将讨论用户体验设计如何影响图像数据的互操作性和重用性。我将通过我对 ImageSnippets (https://imagesnippets.com) 系统的经验来审视这些材料。
This chapter will explore many aspects of the role images can play in PKGs – either as central elements of the design of the PKG or ancillary to the graph. It will also discuss how a user experience design can affect the interoperability and reuse of image data. The material will be examined through the lens of my experiences with the ImageSnippets (https://imagesnippets.com) system.
ImageSnippets 由知识工程师创建，是一个基于 Web 的链接数据注释和元数据管理系统。它以图像为图的中心主题，同时还充当基于图像图的数字资产管理系统。该系统允许非技术用户将图像描述存储为资源描述框架 (RDF) 中的结构化和半结构化机器可读数据。RDF 是用于在 Web 上存储元数据的标准数据模型。
Created by knowledge engineers, ImageSnippets is a web-based linked data annotation and metadata management system. With images as the central subjects of the graph, it also functions as an image-graph based digital asset management system. The system allows nontechnical users to store descriptions of images as structured and semi-structured machine-readable data in Resource Description Framework (RDF). RDF is a standard data model for storing metadata on the web.
十多年来，用户一直在使用 ImageSnippets 描述图像。这引发了大量关于形式化图像描述、元数据以及图像在知识管理过程中如何发挥作用的研究。注释以 JSON-LD 和 RDFa 格式编写在 HTML 文件中，以便共享和发布。它们还保存在具有可公开访问端点的 RDF 数据集中。任何使用端点 URI 的人都可以使用 RDF 的 SPARQL 查询语言查询 ImageSnippets 数据。
For over a decade, users have been describing images with ImageSnippets. This has led to an extensive body of research about formalized image description, metadata, and how images function in the knowledge management process. The annotations are written in JSON-LD and RDFa in HTML files for sharing and publishing. They are also saved in an RDF dataset with a publicly accessible endpoint. Anyone using the URI of the endpoint can query the ImageSnippets data using the SPARQL query language for RDF.
ImageSnippets 系统中有超过 81,000 张关于图像的图表。这些图像来自画廊、图书馆、档案馆和博物馆等各种领域，但该系统也用于描述个人图像收藏。系统中保存的数据（已明确标记为版权的图像除外）也可以描述为 FAIR 数据，遵守 FAIR 原则 [2]，即可查找、可访问、可互操作和可重用。
There are over 81,000 graphs about images in the ImageSnippets system. The images are from a variety of domains such as galleries, libraries, archives and museums, but the system is also used to describe personal image collections. The data saved in the system (except images that have been specifically marked as copyright) can also be described as FAIR data, adhering to the FAIR principles [2] of Findable, Accessible, Interoperable and Reusable.
总的来说，ImageSnippets 系统支持以结构化、机器可读格式捕获的图像描述的精度和细节水平，这是目前任何其他系统（无论是人工还是机器设计的）都无法实现的。
In general, the ImageSnippets system supports a level of precision and detail about image description captured in a structured, machine-readable format that is not currently found in any other system – whether human or machine engineered.
PKG 概念和图像元数据
PKG Concepts and Image Metadata
为了考虑如何将图像以及更重要的是它们的描述集成到个人知识图谱中，我们需要更详细地探索一些基本的图形架构和元数据。
To consider how images and more importantly, their descriptions can be integrated into personal knowledge graphs, we need to explore some basic graphing architecture, and metadata in more detail.
虽然有许多类型的计算架构可用于构建 PKG（包括属性图、markdown 格式、超链接技术和关系数据库结构），但我们将重点关注受 ImageSnippets 系统的研究和创建影响的方法、概念和术语。
While there are many types of computing architectures that can be used in the construction of PKGs (including property graphs, markdown formatting, hyperlinking techniques and relational database structures), we will focus on methods, concepts and terminologies that have specifically been influenced by the research and creation of the ImageSnippets system.
我们从术语“图”开始。在此，我们遵循 RDF 标准 [3]，用于描述基于 Web 的机器可读数据。图由三元组组成，三元组只是三个项目的有序序列。三元组可以被认为是一个非常简单的句子，形式是主语-谓语-宾语，或者更抽象地看作是由一条边连接的两个节点：节点-边-节点。将其画在纸上会得到一个由标记弧连接的节点图，因此有“图”这个术语。虽然单个三元组“句子”并不能表达太多内容，但将它们组合成一个大图可以表达相当复杂的描述。
We start with the term “graph.” In this, we follow the RDF standard [3] for describing web-based machine-readable data. Graphs are made up of triples, which are simply ordered sequences of three items. A triple can be thought of as a very simple sentence of the form subject-predicate-object, or more abstractly as two nodes joined by an edge: node-edge-node. Drawing this out on a page gives a diagram of nodes joined by labeled arcs, hence the “graph” terminology. While a single triple “sentence” does not say much, a lot of them together in a large graph can express quite complicated descriptions.
因此，建设
Thus, the construction
主语 - 谓语 - 宾语
subject - predicate - object
在 RDF 中被称为“三元组”。谓词表示属性（有时也被称为属性），并保存主语和宾语节点之间的关系。
is called a “triple” in RDF. The predicate denotes a property (and is sometimes referred to as a property) and holds a relationship between the subject and object nodes.
在 ImageSnippets 中，关于图像的三元组是关于该图像的命名图，数据集中的所有三元组构成关于系统中所有图像的知识图。此图像图也可以被视为多模态图，因为它包含文本和图像形式的多种模态。
In ImageSnippets, the triples about an image are a named graph about that image and all the triples in the dataset form a knowledge graph about all the images in the system. This image graph can also be thought of as a multimodal graph as it contains a mix of modalities in the form of text and images.
系统中的三元组相互连接，并外部与公共数据集连接，例如 DBpedia（https://dbpedia.org）和 Wikidata（https://wikidata.org）等。
The triples in the system are interlinked to each other and externally to public datasets such as DBpedia (https://dbpedia.org) and Wikidata (https://wikidata.org) among others.
在构建 ImageSnippets 之前，我和来自佛罗里达人类与机器认知研究所 (IHMC) 的知识工程师团队合作开展了一个项目，使用概念图对图像描述中的知识进行建模。[4] 我们的目标是探索艺术家如何描述自己的艺术作品以及如何将这些概念形式化。概念图不是 PKG 表示，但它们可以说明构成 PKG 的一些相同结构，包括以 RDF 语法形式形成的结构。
Prior to the construction of ImageSnippets, myself and a team of knowledge engineers from the Florida Institute for Human and Machine Cognition (IHMC) collaborated on a project to model knowledge found in image descriptions using concept maps. [4] Our goal was to explore how artists described their own artwork and how those concepts could be formalized. Concept maps are not PKG representations, but they can illustrate some of the same structures that make up a PKG including those formed as RDF syntax.
图 9.1 展示了一幅名为“汇合”的艺术照片描述的概念图。[5] 这项研究中的一些想法已被证明与 PKG 的构建有关。其中一个是看似简单的观察，即图像的创作者是他们自己作品的主题专家。这一观察为根据图像创作者的描述数据构建的 PKG 奠定了基础。另一个观察是，我们意识到我们有机会以截然不同的方式研究图像元数据。
Figure 9.1 shows a concept map of a description of a fine art photograph titled, “Confluence.” [5] Several ideas from this research have turned out to be relevant to the construction of PKGs. One was the deceptively simple observation that the creator of the image was the subject matter expert of their own work. This observation established the foundation of what a PKG built from an image creator’s description data might look like. Another observation was that we realized we had an opportunity to study image metadata in distinctly novel ways.
虽然多年来画廊、图书馆、档案馆和博物馆（GLAM 部门）已经使用各种技术处理元数据，但我们希望以全新的视角进行探索。我们的目标不是重新发明已经运行良好的技术，例如国际图像互操作性框架 [6]，而是详细研究图像本身的描述，以寻找可以形式化描述的方法。
While metadata has been addressed in galleries, libraries, archives and museums (the GLAM sector) for many years with an enormous variety of techniques, we wanted to explore it with fresh eyes. Our objective was not to reinvent the wheel of techniques already working well such as the International Image Interoperability Framework, [6] but rather to look in detail at the description of the image itself for ways in which the description could be formalized.
图像元数据可以被认为是关于图像任何部分的任何数据：从图像文件资源本身的技术细节到图像内容的复杂描述。它可以包括图像的创建环境、图像的存储位置以及用于对图像进行分类以便于访问、管理、存档或搜索和搜索引擎优化 (SEO) 的关键字或标题。元数据是图像的“谁？什么？哪里？何时？为什么？以及如何？”。
Image metadata can be thought of as any kind of data about any part of the image: from the technical details of the image file resource itself to complex descriptions of the image contents. It can include the circumstances under which the image was created, where the image is stored and the keywords or captions used to classify the image for accessibility, curation, archiving, or search and search engine optimization (SEO). Metadata is the “Who? What? Where? When? Why? and How?” of the image.
图 9.1 关于标题为“Confluence”的图像的概念图
Figure 9.1 A concept map about the image titled: Confluence
虽然元数据已嵌入数字图像多年，但并不总是得到一致使用。然而，在任何知识图谱中，嵌入元数据的重用都非常有用。根据 PKG 的目标，图谱可能会利用元数据，例如图像的创建日期、图像的主题、图像创建者的姓名或创建图像所使用的技术。它还可以随图像传播到其他环境，同时为图谱提供有用的数据点。
While metadata has been embedded in digital images for many years, it has not always been used consistently. However, the reuse of embedded metadata can be extremely useful in any knowledge graph. Depending on the objectives of a PKG, the graph might utilize metadata such as the creation date of the image, the subject matter of the image, the name of the image creator or the techniques used in the creation of an image. It can also travel with the image to other environments while simultaneously providing useful data points to the graph.
尽管嵌入式元数据具有明显的实用性，但许多应用程序却经常忽略它。虽然 Adobe 和许多照片编辑平台都支持它，但它却一直被社交媒体应用程序忽略或删除。许多当代应用程序要么不知道它的存在，要么声称不知道它的存在。隐私问题经常被认为是删除元数据的原因之一，而且理由充分。相机硬件嵌入为 EXIF 数据的 GPS 位置数据很容易在网上图像中被访问，并被不良行为者用于恶意目的。
Embedded metadata has often been ignored in many applications despite its obvious utility. While supported by Adobe and many photo editing platforms, it has a history of being ignored or stripped by social media applications. Many contemporary applications are either unaware of its existence or claim to be unaware of its existence. Privacy issues are often cited as one reason why metadata is stripped and with good reason. GPS location data embedded as EXIF data by camera hardware can easily be accessed in images online and used with ill intent by bad actors.
对于许多应用程序开发人员来说，隐私和安全的解决方案通常是删除文件中的所有元数据，但对嵌入元数据进行更周到的考虑，可以让许多应用程序受益于至少部分字段的可重用性。允许元数据随图像一起移动也可以鼓励其使用，并用于多模式识别系统。
The solution to privacy and security for many application builders has generally been to remove all the metadata in a file, but a more thoughtful consideration of embedded metadata, could allow many applications to benefit from the reusability of at least some of these fields. Allowing metadata to travel with an image could also encourage its use and be used in multi-modal recognition systems.
然而，对于在 PKG 系统中使用的元数据，应该问这样的问题：
For metadata to be used in a PKG system however, one should ask such questions as:
图像元数据在技术上如何应用于PKG系统？
How can the image metadata be used in the PKG system technically?
无需手动重建元数据，图像及其元数据是否可以在图表中自然地重复使用？
Can the images and their metadata be reused naturally in the graph without manual metadata reconstruction efforts?
关于图像的新输入信息和元数据是否可以在工作流中向前甚至向后传播到先前的图表？
Can newly entered information and metadata about images propagate forwards or even backwards to previous graphs in a workflow?
PKG 用户界面是否应包含图像的元数据编辑功能？
Should the PKG user interface include a metadata editing function for the images?
这些图像及其元数据是否可以在工作流程的下游使用，从而允许在 PKG 之外重复使用这些图像？
Can the images and their metadata be used further downstream in workflows that allows the images to be reused outside of the PKG?
就本章的目的而言，元数据通常可分为两大类：
For the purpose of this chapter, metadata can generally be categorized into two main groups:
1) 主要关于图像或数字文件的创建的元数据：创建者、创建日期和位置（可以存储为 GPS 数据）、版权、相机/扫描仪数据（例如型号、曝光、闪光灯设置等）、文件大小和分辨率等。关于图像中所见项目创建的元数据通常与文件创建的元数据不同。
1) Metadata predominantly about the creation of the image or the digital file: creator, creation date, and location (which can be stored as GPS data), copyright, camera/scanner data such as model, exposure, flash settings etc., file size and resolution, etc. Metadata about the creation of an item seen in an image is often different from the creation of the file.
2）关于图像内容的元数据。这通常包括标题、标题、说明、描述、替代文本和关键字。关键字元数据在不同情况下也可以称为标签、标签、注释或分类。这些标签有很多细微差别，但在本章中，我们大多可以将它们视为同义词。分类通常指机器学习分类器对类别的预测。
2) Metadata about the content of image. This can typically include the title, headline, caption, description, alt-text and keywords. Keyword metadata can also be referred to as tags, labels, annotations or classifications in different circumstances. There are many nuances in these labels, but for this chapter, we can mostly consider them synonymous. Classifications often refer to the predictions of classes made by machine learning classifiers.
第一组中的大部分数据称为 EXIF 数据，通常来自创建原始数字文件的硬件，例如相机、扫描仪和智能手机。[7]
Much of the data in the first group is called EXIF data and usually comes from the hardware that creates the original digital file, such as cameras, scanners and smartphones. [7]
第二组通常嵌入使用国际新闻电信委员会 (IPTC) 标准的文件中。[8] 如今，IPTC 字段采用 Adobe 的可扩展元数据平台 (XMP) 格式化。该标准定义了 56 多个字段。当全部使用时，它可以提供一种相当全面的方法来随图像传输图像描述数据。[9] 图像元数据中使用的其他值得注意的架构包括都柏林核心 (DC) 规范。[10] 一些 DC 字段在 IPTC/XMP 规范中使用。
The second group is usually embedded in files using the International Press Telecommunications Council (IPTC) standards. [8] Today, the IPTC fields are formatted in Adobe’s Extensible Metadata Platform (XMP). There are over 56 fields defined in the standard. When used in its entirety, it can provide a quite comprehensive means of transmitting image description data with images. [9] Other notable schemas used in image metadata include the Dublin Core (DC) specifications. [10] Some DC fields are used in the IPTC/XMP specification.
许多流行的图像编辑和资产管理工具（例如 Photoshop、Lightroom）都包含用于处理 EXIF 和 XMP 数据的元数据编辑面板。所有类型的图像元数据都可以考虑用于 PKG 系统。ImageSnippets 大量使用嵌入式元数据，并且是系统创建的图表的核心。
Many popular image editing and asset management tools, such as Photoshop, Lightroom, include metadata editing panels for working with EXIF and XMP data. All types of image metadata can be considered for use in a PKG system. ImageSnippets makes extensive use of embedded metadata and is central to the graph created by the system.
当考虑一个特定的 IPTC 领域时，允许元数据随图像一起传输的实用性最为明显。替代文本（可访问性）字段于 2021 年添加到 IPTC 标准中。随着日常生活所需的服务越来越多，例如银行、医疗和政府服务几乎完全在线处理，网络内容迫切需要服务于所有人，而不仅仅是视力正常和技术先进的用户。
The utility of allowing metadata to travel with images is most apparent when considering one IPTC field in particular. The alt-text (accessibility) field was added to the IPTC standard in 2021. As more and more services needed for daily life, such as banking, medical and government services are handled almost exclusively online, there is a huge need for web content to serve everyone, not only sighted and technically advanced users.
IPTC 标准中定义的 Alt-Text（可访问性）字段包含的信息与 HTML img 标签规范中 W3C“alt 标签”中的信息相同。自网络诞生以来，alt 属性就一直是 HTML 规范的一部分。它最初旨在包含在互联网早期等待项目加载时可以阅读的描述，当时网页加载速度通常很慢。然而，随着互联网速度的提高，近 20 年来该字段的使用基本上被忽视了。但对于屏幕阅读器软件来说，该属性是一个显而易见的选择，它可以帮助视觉和认知障碍用户读取有关图像的信息。
The Alt-Text (Accessibility) field is defined in the IPTC standard as containing the same information found in the W3C “alt tag” from the HTML img tag specification. The alt attribute has been a part of the HTML specification since the beginning of the web. It was originally intended to contain a description that could be read while waiting for items to load in the early years of the internet when web pages would often be slow to load. When internet speeds increased, however, the use of the field was largely ignored for nearly 20 years. But the attribute was an obvious choice for screen reader software to read information about an image for visually and cognitively impaired users.
Alt-text 与图片的其他标题或描述信息不同。Alt-text 的建议内容由 Web 内容可访问性指南 (WCAG) (https://www.w3.org/WAI/standards-guidelines/wcag/) 定义，并指定应如何向残障网络用户描述网络内容和上下文。
Alt-text is differentiated from other captions or descriptive information about an image. The suggested content for alt-text is defined by the Web Content Accessibility Guidelines (WCAG) (https://www.w3.org/WAI/standards-guidelines/wcag/) and specifies how web content and context should be described for a disabled web user.
但除了让所有人都能更方便地访问网络这一显而易见的目标之外，将此字段添加到嵌入式元数据标准中还激发了网络技术人员的热情，让他们考虑如何支持嵌入式元数据的重复使用。在 PKG 技术中遵守 alt-text 字段也有助于让 PKG 对所有人都更具包容性。
But beyond the obvious goal of making the web more accessible for all, adding this field to the embedded metadata standard has revitalized a fresh enthusiasm by web technologists to consider how to support the reuse of embedded metadata. The observance of the alt-text field in PKG technology could also help make PKGs more inclusive for all.
元数据的主题通常很乏味，但通过在个人知识图谱考虑中包含嵌入式元数据，它可以允许来源、描述和替代文本数据以及许多其他领域用于各种各样的网络体验和工作流程。
The topic of metadata is often tedious, but by embracing embedded metadata in personal knowledge graph considerations, it could allow provenance, descriptive and alt-text data, as well as many other fields, to be available for a wide variety of web experiences and workflows.
从关键词到三重标签
From Keywords to Triple-Tags
我最初的研究是研究描述性元数据的一个非常特殊的方面：关键词。我想探索如何让我自己的图片集里的关键词更清晰、更有意义。
My original research was to examine one very particular aspect of descriptive metadata: the keyword. I wanted to explore how I could make keywords less ambiguous and more meaningful in my own image collections.
关键词或“标签”通常以纯文本字符串的形式包含在数字环境中或与图像一起呈现，或以“主题标签”的形式扩展，主要用于网络和社交媒体系统。关键词的主要用途是帮助在数字资产管理系统中进行图像搜索和检索，以及改善产品营销的搜索引擎优化。标签有时可以在从受控词汇表、分类法和层次结构中提取时嵌套。当关键词围绕语义相似性聚集时，可以创建“标签云”。
Keywords or “tags” are usually included or presented with images in digital environments in the form of plaintext strings or extended in the form “hashtags” predominantly used across the web and in social media systems. The primary use of keywords has been to aid in image search and retrieval in digital asset management systems and to improve SEO for product marketing. Tags can sometimes be nested when they are pulled from controlled vocabularies, taxonomies and hierarchies. “Tag clouds” can be created when keywords are clustered around semantic similarity.
乍一看，可能不太明显，图像上附加的关键字列表可以或应该成为图表的一部分。关键字通常以纯文本字符串的形式存储、传输和表示，并编码为存储在单个字段中的单词或短语的分隔列表。
At first glance, it might not be obvious that keyword lists attached to images could or should become part of a graph. Keywords are often stored, transmitted and represented as plaintext strings and encoded as a delimited list of words or phrases stored in a single field.
不管关键词是由人还是机器推荐的，它们往往由于多种原因而含糊不清且不精确。例如，“cardinal”这个词可能是一只鸟、一个数字系列、天主教会成员或运动队的一名球员，具体取决于上下文。
Regardless of whether keywords are suggested by humans or machines, they are often ambiguous and imprecise for many reasons. The word “cardinal,” for example, could be a bird, a number series, a member of the Catholic Church, or a player on a sports team, depending on context.
但人类和机器都会出于各种“习得”原因创建不相关的标签。人类经常使用冗余标签来尝试改善搜索结果，而机器已经学会了人类创建的标记习惯，以尝试增加图像的搜索结果。
But both humans and machines create irrelevant tags for numerous “learned” reasons. Humans often use redundant tags to try to improve search results and machines have learned the tagging habits humans have created to try to multiply search results for the image.
关键词冗余或目的相悖的例子包括使用“水”和“水体”，然后指定“大西洋”和“海洋”。更智能的系统可以推断出大西洋是一片海洋，是一片水体，是水（与一杯可饮用的 H 2O 不同）。
Examples where keywords are either redundant or at cross purposes include using “water” and “body of water” and then specifying “Atlantic Ocean” and “ocean”. A smarter system can infer that the Atlantic Ocean is an ocean which is a body of water which is water (and not the same kind of water as a glass of drinkable H 2O.)
其他示例包括在关键字列表中使用以下所有关键字：“日落”、“日出”、“黎明”和“黄昏”。而如果人类不熟悉图像的背景而只是猜测，他们也会犯同样的错误。通常，其他线索（例如，如果图像中准确记录了创建日期）可以澄清这种不必要的冗余。
Other examples include using all the following keywords in a keyword list: “sunset,” “sunrise,” “dawn” and “dusk.” While humans can also make this same mistake if they are not familiar with the context of the image and are just guessing. Often, other clues (such as the creation date if recorded in the image accurately) can clarify this needless redundancy.
关键词也经常没有上下文。如果无法明确短语本身与图像内容之间的关系，人们会发现相机型号、技术、事件缩写、人名和隐喻与实际描述场景的概念混杂在一起。
Keywords also often exist without context. Lacking clarification as to how the phrase itself relates to the image content one can find camera models, techniques, acronyms for events, personal names, and metaphors mixed with concepts that are actually describing the scene.
添加尽可能多的关键词，以便可以在许多冗余类别中找到图像，并且对大量图像（特别是在没有上下文和消歧义的情况下）进行大规模搜索，会使搜索结果比必要的更加嘈杂和低效（考虑到现有技术的可能性）。
Adding as many keywords as possible such that images can be found in many redundant categories, to many images at scale – particularly without context and disambiguation – can make search results much more noisy and inefficient than should be necessary – given the possibilities of existing technology.
图 9.2 说明了人类和机器提供的关键字和描述的种类繁多。在这张图片中，对 VisualGPT 服务的调用返回了“停在沙滩上的船”的准确描述，但没有提供其他背景信息。Microsoft Azure 服务将其描述为“沙滩上的木结构”。这张图片来自佛罗里达州档案馆，列出了各种短语，例如船名、建造者姓名和短语“船舶建造”，而 Clarifai 服务和 Azure 服务都返回了一系列预测，主要涉及风暴、腐烂、木材和遗弃的想法，而事实上，这是一艘正在建造的渔船。关键字中列出的名称只有在完整描述提供的上下文中解释时才有意义，完整描述如下：
Figure 9.2 illustrates the overwhelming variety of options for keywords and descriptions supplied by humans and machines. In this image, a call to the service VisualGPT returned an accurate description of a “boat sitting on a beach” but provided no other context. The Microsoft Azure service described it as “wooden structure on a beach.” The image, which comes from the State Archives of Florida listed an assortment of phrases such as the name of the boat, the name of the builders and the phrase, “boat construction,” while both the Clarifai service and the Azure service returned a selection of predictions mostly involving storms, decay, wood and the idea of abandonment when, in fact, this is a fishing vessel under construction. The names listed in the keywords only make sense when interpreted with the context provided by the full description which was supplied as:
“建造期间从右舷看 Miss Joann 号船头：佛罗里达州梅波特，1985 年 7 月 13 日”
“Bow of the Miss Joann from the starboard side during construction: Mayport, Florida July 13, 1985”
附注：“查尔斯、唐纳德和托马斯·赫林兄弟在佛罗里达州梅波特建造了 54 英尺长的虾拖网渔船“乔安小姐”号。”
Accompanying note: “Construction of 54-foot shrimp trawler, Miss Joann, by brothers Charles, Donald, and Thomas Herrin, at Mayport, Florida.”
图 9.2 建造中的渔船图像，周围是图像创建者、存档服务和对多个 AI 服务的调用提供的各种关键字和描述。
Figure 9.2 An image of a fishing vessel during construction surrounded by a variety of keywords and descriptions supplied by the image creator, the archiving service and calls to several AI services.
当我们的研究团队开始使用概念图分析图像关键词时，可能会使用 RDF 三元组代替关键词，这给了我们一个绝佳的机会来更深入地思考这些模糊的分类，并以全新的眼光看待它们。我们并不满足于仅仅创建一个称为“关键词”的关系，用一个充满纯文本字符串的节点，而是更进一步，开始使用语义网中的哲学作为指导来评估关键词和概念。
When our research team began analyzing image keywords using concept maps with the potential of using RDF triples in place of keywords, it gave us an excellent opportunity to think much more deeply about these ambiguous classifications and see them in a new light. Instead of being satisfied with merely creating just one relationship called “keywords,” with a node filled with plaintext strings, we went a step further and began evaluating keywords and concepts using philosophies from the semantic web as a guide.
考虑一下，对于简单的关键字“树”，可能会返回多少张不同的图片。即使同时使用三个关键字，例如“树、鸟、水”，也可能会返回一组完全不同的、嘈杂的结果。这在大量图片中尤其明显，因为对于这些关键字，可能会返回数千张图片。
Consider how many different images could be returned for the simple keyword “tree.” Even using three keywords together such as “tree, bird, water” could return a disparate, noisy set of results. This is particularly obvious in a large set of images where thousands of images might be returned for these keywords.
在我们的工作中，我们的假设是，通过将图像本身或图像的某个区域分配给 RDF 三元组的主体位置、将关键字分配给客体位置、将关系或上下文分配给谓词位置，我们可以创建“三重标签”来提高图像的可查找性。
In our work, our hypothesis was that by assigning the image itself or a region of the image to the subject position of the RDF triple, the keyword to object position, and a relationship or context to the predicate position, we could create “triple-tags” that would improve the findability of images.
链接数据是一组用于连接已存储在 RDF 表示中的数据的最佳实践。我们的理论是，我们可以使用链接数据来搜索公共数据集中的实体，它可以像受控的分层词汇表一样发挥作用。通过使用传递属性和推理，我们还可以使用简单知识组织系统 (SKOS) 概念和子类关系（例如 DBpedia 和 Wikidata 中发现的那些）来增加跨主题领域建立更多连接的潜力。
Linked data is a set of best practices for connecting to data that has been stored in RDF representations. Our theory was that we could use linked data to search for entities in public datasets and it could act in the same way as controlled hierarchical vocabularies. By using transitive properties and inference we could also add the potential for many more connections across topic areas using the Simple Knowledge Organization System (SKOS) concepts and subclass relationships such as those found in DBpedia and Wikidata.
因此，例如，只需为描绘“鹈鹕”的图像保存一个三重标签，就可以在搜索“鸟”时找到该图像，而无需重复使用“鸟”关键字。SKOS 概念还可以让图像在各种搜索中找到，包括更抽象的概念，例如：“巴巴多斯的国家象征”，其中“鹈鹕”就是其中之一。
Thus saving just one triple-tag for an image depicting a “pelican,” for example, would allow that image to be found in a search for “bird” without the need of the redundant use of the “bird” keyword. SKOS concepts would also allow the image to be found for a variety of searches including much more abstract concepts such as: “National symbols of Barbados,” of which “pelican” is one.
链接数据解决了消除关键词本身歧义的重要问题，但我们持续研究中最具挑战性的部分是识别三元组的谓词部分的关系，这种关系可以定义关键词/对象与图像/主题的关系。
Linked data solved the important problem of disambiguating the keyword itself, but the most challenging part of our continued research was in identifying the relationships for the predicate part of the triple that could define the way in which the keyword/object would relate to the image/subject.
在考虑如何处理三重标签的关系部分时，我们首先查阅了原始的基于 RDF 的朋友之友词汇表 (FOAF)。[11] FOAF 词汇表提供了“描述”的概念作为图像的关系。
When considering how to treat the relationship part of our triple tag, we first consulted the original RDF based Friend-of-a-Friend vocabulary (FOAF). [11] The FOAF vocabulary offered the concept of “depiction” as a relationship for images.
然而，很快就发现并非所有用关键词描述的对象都是图像中“描绘”的事物。检查图 9.2 中的关键词可以很好地说明这一点。那么问题是：我们应该使用多少关系以及如何表示它们？试图解析描述中每个句子中的每个谓词会产生太多关系 - 其中许多关系在语义上是同义词，难以记住，并且在自然语言中语法过于复杂，无法用作 RDF 中的谓词。
However, it was quickly apparent that not all objects described by keywords are things that are “depicted” in images. Examining the keywords in figure 9.2 provides a good illustration of this. The question was then: how many relationships should we use and how should they be represented? Trying to parse out every predicate in every sentence in a description yields way too many relationships – many of which are semantically synonymous, difficult to remember and too grammatically complex in natural language to be useful as predicates in RDF.
我们的概念图工作帮助我们识别了风景、图画和解释元素等区别，但决定创建一个“三重标签”编辑器（现在是 ImageSnippets 系统的核心功能之一）使我们能够应用实际研究来寻找如何最好地解决三重标签中的谓词值。
Our concept mapping work helped us identify distinctions such as scenic, pictorial and interpretive elements, but deciding to create a “triple-tag” editor – which is now one of the core functions of the ImageSnippets system – allowed us to apply hands-on research for how best address the predicate value in the triple tags.
我们的关系词汇量随着时间的推移不断增长，形成了一种小型本体，称为轻量级图像本体 (LIO)。这里的“本体”一词是指嵌入知识图谱中的相关概念的集合，它定义了这些概念的预期含义。
Our vocabulary of relationships grew over time took the form of a small ontology called the Lightweight Image Ontology (LIO). The word “ontology” here means a collection of related concepts embedded in a knowledge graph which defines their intended meaning.
LIO 的创建是一个反复的过程，用户可以自由地创建图像标记。然后，本体论者检查标记，并在必要和实用的地方扩展词汇表。在整个过程中，我们立即认识到在概念记录过程中可以捕捉知识的精确时刻的价值和重要性。我们将这一时刻称为信息记录点，即图像创建者或注释者有话要说并准备好说出来的时刻。[12]
The creation of LIO was the result of an iterative process where users could create image markup freely. The markup was then examined by ontologists and the vocabulary extended where necessary and pragmatic. Throughout this process we immediately recognized the value and importance of the precise moment in which the knowledge could be captured in the conceptual recording process. We refer to this moment as the information recording point which is the moment that the image creator or annotator has something to say and is ready to say it. [12]
图 9.3 在 ImageSnippets 中的注释编辑器中注释的 Confluence 图像的屏幕截图。
Figure 9.3 A screenshot of the Confluence image being annotated in the annotation editor in ImageSnippets.
图 9.3 显示了三元组编辑器的运行情况。用户已从图 9.1 中的 Confluence 图像中选择了一个区域。用户可以选择图像或图像的某个区域作为三元组的主题，然后从 LIO 中选择关系（或创建新谓词），然后从链接数据数据集中快速查找匹配的实体。在本例中，用户从 DBpedia（https://dbpedia.org/page/Wood_grain）中选择了“木纹”。
Figure 9.3 shows the triple editor in action. A user has selected a region in the Confluence image from figure 9.1. The user can select the image or a region of the image as subject of the triple, then choose a relationship from LIO (or create a new predicate) and then quickly perform a lookup for a matching entity from a linked data dataset. In this case the user has selected “wood grain” from DBpedia (https://dbpedia.org/page/Wood_grain).
用户只需几秒钟即可在界面上创建此注释。使用这一个三重标签，现在可以使用 DBpedia 中的 SKOS 搜索“木材、木工和木雕”以及使用 Wikidata 中的术语“纹理图案”来找到此图像。
It took just seconds for the user to create this annotation in the interface. Using this one triple tag, this image can now be found in a search for: wood, woodworking and woodcarving using SKOS in DBpedia and with the term “grain pattern” from Wikidata.
因为 DBpedia:Wood_grain 与“木纹”的 Wikidata 实体相同，所以 sameAs 关系映射了两个数据集之间的等效类，并允许在具有更多术语的系统中找到图像。
Because DBpedia:Wood_grain is the same as the Wikidata entity for “Wood Grain.” the sameAs relationship maps the equivalent classes between the two datasets and allows the image to be found in the system with more terms.
生成的注释也被写成 JSON-LD 和 RDFa 形式存在于有关图像的 HTML 文件中，并在此处显示为 RDFa。
The resulting annotation is also written out as JSON-LD and RDFa in an HTML file about the image and shown here as RDFa.
<span about="https://imagesnippets.com/imgtag/images/info@margaretwarren.us/confluence1.jpg"><span rel="lio:hasVisualPart"><span resource="#Region_Region%20A"></span></span></span>
<span about="https://imagesnippets.com/imgtag/images/info@margaretwarren.us/confluence1.jpg"><span rel="lio:hasVisualPart"><span resource="#Region_Region%20A"></span></span></span>
<span about="#Region_Region%20A"><span rel="lio:shows"><span resource="[dbpedia:Wood_grain]"><span property="rdfs:label" content="木纹"></span></span></span></span></span>
<span about="#Region_Region%20A"><span rel="lio:shows"><span resource="[dbpedia:Wood_grain]"><span property="rdfs:label" content="wood grain"></span></span></span></span>
LIO 词汇表 [13] 包含 11 个属性，自 2013 年以来一直保持稳定，并被许多未接受过 RDF 或语义网技术培训的注释者使用。LIO 本体在图 9.4 中以直观方式说明，可以在 https://imagesnippets.com/ArtSpeak/help/properties.html 中以扩展形式查看。
The LIO vocabulary, [13] comprising 11 properties, has been stable since 2013 and used by numerous annotators not trained in RDF or semantic web technologies. The LIO ontology is visually illustrated in figure 9.4 and can be seen in an expanded form at https://imagesnippets.com/ArtSpeak/help/properties.html.
图 9.4 LIO 属性的可视化示例
Figure 9.4 Visual examples of the LIO properties
链接数据注释的实践
Linked Data Annotations in Practice
自三重标签编辑器创建以来的十年里，我们的团队已经对以三重标签表示的关键词的价值以及生成的 RDF 图的实用性进行了大量的研究。
In the decade since the creation of the triple-tag editor, our team has since been able to conduct much research about the value of keywords represented as triple-tags and the utility of the resulting RDF graph.
下面的示例是对使用 ImageSnippets 三重标签编辑器构建的图像描述的完整注释处理的分析。
The following example is an analysis of a complete annotation treatment of an image description constructed with the ImageSnippets triple-tag editor.
在这个例子中，一位熟悉 ImageSnippets 的熟练注释者检查了图像创建者以及各种机器学习服务创建的图像描述和元数据。然后，注释者将所有这些数据转换成三重标签。值得注意的是，虽然这个分析似乎表明这个过程很耗时，但这是一个典型的图像示例，注释者花费的时间与创建传统关键字列表的时间大致相同。图 9.6 中显示的 15 个三重标签大约在两到三分钟内创建完成。
In this example, a skilled annotator familiar with ImageSnippets, examined the image description and metadata created by both the image creator as well as various machine learning services. The annotator then translated all this data into triple-tags. It should be noted that while this analysis would seem to suggest that the process was time consuming, this is a typical example of an image that took the annotator approximately the same amount of time it would have taken for a traditional keyword list to be created. The 15 triple-tags shown in the figure 9.6 were created in about two to three minutes.
图 9.5 泥浆中的船的图像
Figure 9.5 An image of a boat in the mud
图 9.5 是一张英国泰晤士河附近泥泞中的小船图片。该图片拥有 Creative Commons CC-BY 2.0 许可。照片来源：Les Chatfield (https://www.flickr.com/people/elsie/) http://imgsnp.co/mbns6
Figure 9.5 is an image of a boat sitting in the mud near the River Thames in the UK. The image has a Creative Commons CC-BY 2.0 license. Photo credit goes to: Les Chatfield (https://www.flickr.com/people/elsie/) http://imgsnp.co/mbns6
请注意，此图像的解释与图 9.2 中的图像非常相似。船只的大小、状况和位置明显不同，但机器生成的描述大致相同。
Note that this image could also be interpreted very similarly as the image in figure 9.2. The boats are clearly different in size, condition and location, but the machine generated descriptions were much the same.
这幅图的标题是船的名字“古城”，图片创作者这样描述他的标题：
The title of this image is the name of the boat, “Ancient City,” and the image creator described his title as:
“这艘在泥浆中慢慢腐烂的旧船有着一个有趣的名字。”
“The intriguing name given to this old launch gently mouldering away on the mud.”
从而更准确地识别船只的类型。
which identified the type of boat more precisely.
图片创建者添加了以下关键字：
The image creator added the following keywords:
银行、 鸟、 鸟类、 船、 羽毛、 飞、 人行道、 莫名的忧郁、 泥浆、 老、 河、 搁浅在绿地上、 太阳、 泰晤士河、 树木、 水
bank, bird, birds, boat, feather, fly, footpath, ineffable melancholy, mud, old, river, strand on the green, sun, Thames, trees, water
ImageSnippets 三重编辑器允许用户调用各种 AI 服务来提供字幕和关键字建议。对第三方机器学习分类器服务 Clarifai [14] 的 API 调用返回了以下预测分类：
The ImageSnippets triple-editor allows users to make calls to a variety of AI services for captioning and keyword suggestions. An API call to the third-party machine learning classifier service Clarifai [14] returned the following predicted classification:
水, 船, 海滩, 运输系统, 船只, 划艇, 无人, 海, 海滨, 旅行, 砂, 车辆, 海洋, 自然, 旧, 河, 弃, 户外, 独木舟, 木
water, boat, beach, transportation system, watercraft, rowboat, no person, sea, seashore, travel, sand, vehicle, ocean, nature, old, river, abandoned, outdoors, canoe, wood
对 Microsoft Azure 服务 [15] 的调用返回了以下内容：
A call to the Microsoft Azure service [15] returned the following items:
标题为“海滩上的一艘船”，虽然总体来说不错，但却遗漏了许多背景信息。
the caption: “a boat on the beach” on the beach, which, while generically good – leaves out much context.
这是一条河边泥泞中腐烂的废弃小船。从技术上讲，“海滩”一词在与河流联系时可能可以接受，但很可能不会在实际中用于此处。
It’s a rotting, abandoned, boat in the mud next to a river. The use of the word “beach” might be technically acceptable in association with a river, but most likely not used in practice at this location.
分类词：户外、交通、地面、船舶、水、船、湖、轮船、独木舟、海滩、坐着、岸边、脏、沙子
and the classifiers: outdoor, transport, ground, watercraft, water, boat, lake, ship, canoe, beach, sitting, shore, dirty, sand
Azure 中的物体检测算法也能够将该物体识别为船。
The object detection algorithm in Azure was also able to identify the object as a boat.
虽然机器学习系统生成了相当好的术语，但它们都缺乏上下文并且包含冗余。
While the machine learning systems generated fairly good terms, they both lacked context and contained redundancy.
总的来说，这些例子说明，当图像内容可以被绘制成更精确、更相关的数据结构时，我们将看到更少的标签可以产生相同或更好的搜索结果。
In general, these examples illustrate that when image content can be graphed into much more precise and relevant data structures, we will see that fewer tags can yield the same or better search results.
例如，可以在 DBpedia/Wikidata 中找到实体“launch”，其定义为：“Launch 是几种不同类型船只的名称。该名称的使用范围很广，从实用船只到按照非常高的标准建造的游艇。”
For example, the entity, “launch” can be found in DBpedia/Wikidata and is defined as: “Launch is a name given to several different types of boats. The wide range of usage of the name extends from utilitarian craft through to pleasure boats built to a very high standard.”
因为我可以将实体“发射（船）”与 DBpedia 实体一起使用，所以它将 DBpedia SKOS 概念和 Wikidata 子类与“船”、“船只”和“摩托艇”等术语相关联。
Because I can use the entity, “launch (boat)” with a DBpedia entity, it is then related DBpedia SKOS concepts and Wikidata subclasses to terms for “boat,” “watercraft,” and “motorboat” among others.
可能与摄影师的经历有关但与图片的实际内容无关的项目包括鸟、鸟类、羽毛、苍蝇、人行道、太阳和树木，因为这些都无法在图片中看到。短语“难以言喻的忧郁”标识了一个 Flickr 群组。
Among the items that may have been relevant to the photographer’s experience, but not to the actual contents of the image were Bird, Birds, Feather, Fly, Footpath, Sun, and Trees as none of these can be seen in the image. The phrase “Ineffable melancholy” was identifying a Flickr group.
表 9.1 是 Clarifai 服务和 Microsoft Azure 服务生成的机器标签列表以及与更精确的注释方法相比它们的实用性的分析，而图 9.6 是一个表格，显示了这些关键词如何转换为三重标签并作为实体存储在图中。
The table 9.1 is the list of the machine tags generated by both the Clarifai service and the Microsoft Azure service and an analysis of the utility they have when compared to a more precise annotation method while figure 9.6 is a table showing how these keywords were transformed to triple-tags and stored as entities in the graph.
表 9.1
Table 9.1
图 9.6 图 9.5 中的图像的三重标签
Figure 9.6 Triple-tags for the image in figure 9.5
这艘船被称为“小艇”，使用了创作者自己的名字，并对其状况进行了多次三元组描述。注释者还创建了一个未被创作者或人工智能注意到的额外实体，用于识别船的明显肋骨，并使用了 DBpedia 中的一个实体来描述航海意义上的船肋骨。甚至在关键词“难以言喻的忧郁”中找到的 Flickr 群组名称也在某种程度上被保留了下来，其中有一个三元组描述了图像传达的忧郁感。
The boat was called a “launch” using the creator’s own name for it and several triples were made about its condition. The annotator also created an extra entity not noted by the creator or AI that identified the noticeable ribs of the boat and used an entity from DBpedia that describes the ribs of a boat in the nautical sense. Even the Flickr group name found in the keywords “Ineffable Melancholy” was preserved in some sense with a triple describing that a sense of melancholy was conveyed by the image.
将这张图片的关键词和描述转换为三重标签的做法在 ImageSnippets 中非常常见，因为我们不是图片的创建者。我们确实有创建者提供的详尽描述，可以从中获取如何构建三重标签的线索。在这种情况下，图片创建者是其图片的主题专家。显然，有些图片比其他图片更容易描述，这取决于注释者有多少非结构化数据可用。如果图片创建者知道如何自己构建三重标签以获得最佳结果，也会有所帮助。在许多情况下，我们努力将最有价值和最相关的关键词的子集转换为三重标签，从而减少构建三重标签所需的时间。我们当前的研究通常涉及研究与纯文本关键词相比，从三重标签中获得的精度类型中究竟在何处和何时可以获得最大价值。此注释处理说明了一种工作流程，其中主题专家（在本例中为图像创建者）提供了描述和关键字建议，而熟练的注释者创建了三重标签作为服务。
The transformation of this image’s keywords and description to triple-tags was fairly typical of many images we have worked with in ImageSnippets where we were not the creator of the image. We did have a thorough description provided by the creator from which to use for clues in how to build the triple-tags. In this case, the image creator was the subject matter expert of their image. Obviously, some images are easier to describe than others and it depends on how much unstructured data is available to the annotator. It can also help if an image creator knows how to construct triple-tags themselves for the best results. In many cases, we strive to convert a subset of the most valuable and relevant keywords into triple-tags, which decreases the time required to build the triple tags. Our current research often involves studying exactly where and when the most value can be derived from the type of precision one can get from triple-tags when compared to plaintext keywords. This annotation treatment illustrates a workflow in which a subject matter expert (in this case, the image creator) provided a description and keyword suggestions and a skilled annotator created the triple-tags as a service.
图像知识工程
Knowledge Engineering with Images
继续前面的例子，从图像中绘制个人知识的哲学可以得到扩展。当图像是图表的核心和描述过程的起点时，图像可能会自然而然地刺激创作者的认知过程。图像本身成为将更多知识带到表面的助记符。如上一节所述，图像创建者（或主题专家）通常对图像的背景和内容最为了解。如果创作者对其主题了解很多，那么后续描述甚至可以采用专家领域知识的形式。创作者自己在信息记录点创建三重标签的难易程度也会影响图表的实用性。
Continuing with the previous example, the philosophy of graphing personal knowledge from images can be expanded upon. When images are central to the graph and the starting point of the descriptive process, images might naturally stimulate the cognitive process for the creator. The image itself becomes the mnemonic aid that brings more knowledge to the surface. As noted in the last section, it is the image creator (or a subject matter expert) who often has the most knowledge of the context and contents of an image. If the creator knows a lot about their subject, the subsequent descriptions could even take the form of expert domain knowledge. The ease with which a creator themselves can create the triple-tags at the information recording point can also influence the utility of the graph.
相反的例子是，当个人知识图谱纯粹围绕基于文本的概念构建时，图像可能很有用，但图像是事后才附加的。当图像只是图中节点的图示时，大量信息可能会丢失或永远无法捕获。
The opposite example is when personal knowledge graphs are built purely around text-based concepts in which images could be useful, but the images are attached as an afterthought. When an image is only an illustration of a node in a graph, a great deal of information might be lost or never captured.
想象一下，图 9.5 中船只图像的创作者对这个主题了解更多。也许他研究过这艘船，并且对其所有权、历史或其他建造细节有更多了解。也许他拍摄过许多其他类似的船只，或者在很多年和很多季节里拍摄过这艘船。他可能从这个位置拍摄了许多不同角度、光线、天气或其他条件的图像。
Imagine that the creator of the boat image in figure 9.5 knew more about this subject. Perhaps he had researched the vessel and knew more about its ownership, history, or other construction details. Perhaps he had photographed many other boats like it or photographed this boat throughout many years and seasons. He might have many images from this location with different angles, lighting, weather or other conditions.
他的个人图像收藏不仅可以保证对这些材料的更深入处理，还可以围绕他对这个主题的深入了解创建自定义词汇表。事实上，可以为这艘真正的船创建一个实体“古城”，并为船的各个部分创建子类关系。自定义谓词还可以帮助他描述有关船的更多属性。该图还可以与其他可能对此类船只有更多领域知识的合作者一起扩展，他们可以为该用户的知识做出贡献。
Not only could his personal image collection warrant an even deeper treatment of this material, but a custom vocabulary could also be created around his deep knowledge of this subject. In fact, an entity could be created for this actual boat, “Ancient City,” and subclass relationships created for the boat parts. Custom predicates could also help him describe more attributes about the boat as. The graph could also be extended with other collaborators who may know even more domain knowledge about these kinds of vessels who could contribute to this user’s knowledge.
构建三元组的努力通常还可以激发潜在遥远概念之间的联系，并阐明人们对其主题的真正了解程度的差距。在这种情况下，使用界面的摩擦实际上可能是有益的，因为绘制具有挑战性的概念通常可以让图像创建者认识到他们知识模型中的不一致之处。
The effort of constructing the triples can also often stimulate connections between potentially distant concepts and illuminate gaps in what someone truly knows about their subject. In this case, the friction of using the interface can actually be helpful because graphing challenging concepts often allows image creators to recognize inconsistencies in their knowledge model.
从个人知识中提取领域专业知识
Extracting Domain Expertise from Personal Knowledge
以下示例是由一位一直在修复自己拥有的汽车并对老式保时捷 356 汽车领域有特别了解的人创建的图像。显示的图像是汽车保险杠的环绕边缘。
The following example is an image created by someone who has been restoring a car they own and has particular knowledge of the vintage Porsche 356 automobile domain. The image shown is a wraparound edge of a bumper on a car.
图 9.6 图 9.5 中的图像的三重标签
Figure 9.6 Triple-tags for the image in figure 9.5
该图片来自专业汽车修复师的图片集，因此创作者了解汽车和保险杠的许多细节，而普通观察者则不了解这些细节。通过以该图片为起点，我们识别出了许多关于早期保时捷保险杠样式的概念，而这些概念可能无法在仅基于文本的制图工具中回忆起来。
The image is from an image collection of a professional car restorer, so the creator knew many details about the car and the bumper, not obvious to the casual observer. By using the image as a starting point, numerous concepts about early Porsche bumper styles were identified that might not have been recalled in a graphing tool that was only text based.
他们的图像描述是：
Their image description was:
“这张图片展示了早期四位数 VIN 保时捷 356 Pre-A 车型上所谓的‘车身保险杠’的边缘。这种特殊的保险杠样式是早期保时捷 356 车型所独有的。它的主要特点是它遵循车身线条，环绕挡泥板，具有光滑的装饰和铝制饰边。饰边已卷绕在保险杠边缘。”
“This image shows the edge of what’s called a ‘body bumper’ on a very early 4-digit VIN Porsche 356 Pre-A model. This particular bumper style is unique to the early Porsche 356 models. Its defining characteristics are that it follows the line of the body, wraps around the fenders, has a smooth deco, and aluminum trim. The trim has been rolled around the edge of the bumper.”
创作者还知道车辆的车辆识别号 (VIN)、前任车主是谁、最初从工厂交付的地点，以及有关该车的许多其他历史和修复相关细节。查看早期保时捷汽车保险杠的许多照片实际上可能有助于其他保时捷修复者完善他们对早期保时捷正确的“车身保险杠”形状的理解。
The creator also knew the Vehicle Identification Number (VIN) of the vehicle, who its previous owners were, where it was originally delivered from the factory, as well as many other historical and restoration related details about the car. Examining many photos of bumpers from early Porsches might actually help other Porsche restorers refine their understanding of exactly what a correct “body bumper” shape on an early Porsche should look like.
此示例与之前示例的主要区别在于，此图像非常难以描述。对于机器来说，这也是一张难以解释的图像，因为没有机器学习模型接受过关于保险杠的训练。描述它还需要大量的领域专业知识和背景。Azure 服务将此图像描述为：“轮胎的特写”。
The chief difference of this example to our previous examples is that this image is quite difficult to describe. It is also a difficult image for a machine to interpret, as no machine learning models have been trained on bumpers. It also requires a lot of domain expertise and context to describe. The Azure service described this image as: “a close up of a tire.”
此外，虽然 DBpedia 和 Wikidata 包含“保险杠”实体，但它们不包含汽车上不同保险杠样式的具体细节，也不包含它们的特性与早期保时捷 356 车型的关系。使用 ImageSnippets 中的三重标签编辑器，注释者可以创建其他数据集中不存在的实体，但这可能很耗时，并且需要付出很多努力才能构建对领域有价值的高质量子类关系。
Also, while DBpedia and Wikidata contain entities for “bumpers,” they don’t contain the particular details of different bumper styles on automobiles nor how their characteristics relate to early Porsche 356 models. Using the triple-tag editor in ImageSnippets, an annotator can create entities that do not exist in other datasets, but this can be time consuming and require much effort to construct good quality subclass relationships that would be valuable to the domain.
但是，如果该领域具有足够的价值，本体工程师可以从这些描述中获取尽可能多的原始元数据，并从图像创建者描述的概念和关系中形成更精确的知识表示形式。从图像描述中得出的有关早期保时捷汽车的潜在有用关系可能是：
However, if there were enough value in the domain, an ontology engineer could take as much of the raw metadata as possible from descriptions such as these and form a much more precise representation of the knowledge into an ontology from the concepts and relationships described by the image creator. Potentially useful relationships about early Porsche automobiles derived from the image description could be:
保时捷 356 Pre-A 车型的 VIN 为 4 位数字
A Pre-A Model of Porsche 356 has a 4-digit VIN
4位VIN的车和Pre-A型号356一样
A 4-digit VIN car is the same as Pre-A model 356
车身保险杠风格是 Pre-A 保时捷 356 独有的
A Body Bumper Style is unique to Pre-A Porsche 356
车身保险杠的样式遵循车身的（几何）线条
The Body Bumper style follows the (geometric) line of the body
平滑装饰是 Body Bumper 风格的一个特点
Smooth Deco is a characteristic of the Body Bumper style
铝制饰条是车身保险杠风格的一个特征
Aluminum Trim is a characteristic of the Body Bumper style
铝制饰条附加特性：卷绕在保险杠边缘
Aluminum Trim additional characteristic: rolled around edge of bumper
一旦围绕该领域创建了词汇表，专家知识就可以成为协作知识工程环境的基础。这样，个人知识图谱就可以成为集体图谱的一部分，所有用户和主题专家都可以贡献他们的专业知识。在对图像进行语义搜索时，用户可以互相帮助，更清楚地校准自己对概念的想法，而对类似图像的检查也有助于为概念建立基本事实。
Once a vocabulary had been created around the domain, the expert knowledge could form the basis of a collaborative knowledge-engineering environment. In this way, the personal knowledge graph could become part of a collective graph where all users and subject matter experts could contribute their expertise. When doing semantic searches for images, users could help each other calibrate their own ideas about their concepts more clearly and the examination of similar images could also help establish ground truths for concepts.
图像图形和互操作性
Image Graphs and Interoperability
考虑到可以存储在图像中并附加到图像中以供重复使用的信息量，这些数据中有多少可以重复使用取决于应用程序中如何使用图像或外部链接。
Considering the wealth of information that can be stored in, and attached to, images for reuse, how much of this data can be reused depends on how images or external links are used in the application.
并非所有 PKG 系统都可以轻松交换图像元数据和注释。例如，使用 markdown 格式的 PKG 通常使用 HTML img 标签处理图像，但唯一的元数据潜力是通过 img 标签中可用的属性。虽然嵌入的元数据可能不会从图像中剥离，但它不太可能在应用程序中提供。ImageSnippets 确实提供了一种通过 URL 共享图像的方法，可以在其中查看图像，并且图像包含嵌入的元数据和以多种格式和模式编写的结构化数据，但许多应用程序仅设计为将图像作为图像文件附加到记录，而不是以链接数据或 RDF 表示的 URL 或 URI。
Not all PKG systems can interchange image metadata and annotations easily. PKGs using markdown formats, for example, typically handle images with the HTML img tag, but the only metadata potential is through the attributes available in the img tag. While the embedded metadata might not be stripped from an image, it’s not likely that it would be made available in the application. ImageSnippets does provide a way for images to be shared via URLs where the image can be viewed and in which the image contains the embedded metadata and the structured data written in several formats and schemas, but many applications are designed to only attach the image to records as image files, not URLs or URIs expressed as linked data or RDF.
一些应用程序或系统将图像作为用户体验的一部分，但无法最大限度地发挥图像描述数据的潜力。例如，Ancestry.com 等 Web 应用程序以人类姓名树的结构为中心，用户可以在其中将姓氏联系在一起。图像可以作为支持材料附加到树上。照片中的人可以与家谱中的人联系起来。Ancestry.com 不会剥离嵌入的元数据，而是提示用户重新输入数据以将其链接到他们的数据库结构。
Some applications or systems include images as part of the user experience but can’t maximize the potential for the image description data. A web application such as Ancestry.com, for example, is centric to the structure of a tree of human names in which users can link family names together. Images can be attached to the tree as supporting material. People who are in photos can be linked to people in the family tree. Ancestry.com does not strip the embedded metadata but prompts the user to reenter data to link it to their database structure.
但请考虑一下，系统不仅可以重复使用嵌入的元数据，还可以重复使用与每幅图像相关的链接数据注释。链接数据注释可以从链接数据生态系统中提供语义上更加丰富的体验。
But consider the idea that systems could not only reuse embedded metadata but also reuse the linked data annotations connected to each image. The linked data annotations could provide a much more semantically enriched experience from the linked data ecosystem.
因为 RDF 是用 URI 构建的，所以系统可以指向任何图像 URI 作为主题，而不必将图像本身存储在 ImageSnippets 服务器上，尽管嵌入的元数据仅当系统对图像具有写访问权限时才可以存储在图像中。ImageSnippets 中的每个图像（或图像中的某个区域）都是图中每个三元组的主要主题（“顶部”、“中心”或“根”）节点。图像也是设计的核心元素，因此它被赋予了头等地位，就像图像在数字资产管理系统中被赋予头等地位一样。关于图像的所有三元组都指向回图像。生成的图像图是一个多模式以图像为中心的图，具有孤立星形的拓扑结构，但有少数例外，其中之一是三元组可以在系统内将图像相互连接。
Because RDF is constructed with URIs, the system can point to any image URI as a subject and does not have to store an image itself on the ImageSnippets server, although the embedded metadata can only be stored in an image if the system has write access back to the image. Every image (or a region in the image) in ImageSnippets is the main subject (“top,” “center” or “root”) node of every triple in the graph. The image is also the central element of the design and because of this it is given first-class status in much the same way an image is given first-class status in a digital asset management system. All the triples about an image point back to the image. The resulting image graph is a multimodal image-centric graph with a topology of isolated stars with few exceptions, one being that triples can connect images to each other within the system.
ImageSnippets 可以提供逆向关系以便在其他 RDF 类型系统中重用，但是为了使以图像为中心的链接数据注释实现与其他系统的真正互操作性，需要考虑所有环境中的数据模型、应用程序架构、用户体验和工作流程。
ImageSnippets can provide inverse relations for reuse in other RDF type systems, but for image-centric, linked-data annotations to achieve true interoperability with other systems, considerations need to be made in the data models, application architecture, user experiences and workflows across all environments.
关联数据中的推理和不一致性
Inference and Inconsistencies in Linked Data
人类擅长的一件事是能够快速发现查询返回的不符合预期结果的图像。虽然机器可以根据数据或视觉相似性返回搜索图像，但它们不知道图像何时与查询不匹配或显示不理想的结果。这种现象最明显的是，计算机视觉返回的图像人类一眼就能看出是刻板印象或明显错误，比如将松饼与吉娃娃混淆。一些机器学习系统用于异常检测，但图像结果的不一致通常非常微妙。人类天生更善于解释为什么图像不适合作为查询的有效结果。
One of the things that humans can excel at is very quickly spotting images that have been returned for a query that do not fit the expected results. While machines can return images for searches based on data or visual similarity, they have no idea when images don’t match that query well or illustrate undesired results. This phenomenon is most obvious when computer vision returns images that humans recognize immediately as stereotypes or obviously wrong in hilarious ways, such as confusing muffins with chihuahuas. Some machine-learning systems are used for anomaly detection, but inconsistency in image results is often very subtle. Humans are naturally better at interpreting why images do not fit as a valid result for a query.
在 ImageSnippets 中，当将三重标签添加到 RDF 数据集时，会为链接数据集中找到的所有传递路径保存一个推理图。这个推理图使我们能够使用来自 SKOS 的类别和来自 Wikidata 的子类来搜索图像。我们建立了一个内部最佳实践，即首先在 DBpedia 中解析实体，尽管 DBpedia 和 Wikidata 经常包含重复的实体。这是因为 DBpedia 通过特殊的 :sameAs 关系链接到 Wikidata，我们能够最大限度地提高语义搜索结果的可能性。
In ImageSnippets an inference graph is saved for all the transitive paths found in linked datasets when triple-tags are added to the RDF dataset. This inference graph is what gives us the ability to search for images using categories from SKOS and subclasses from Wikidata. We established an internal best practice of resolving entities first in DBpedia even though both DBpedia and Wikidata often contain duplicate entities. This is because DBpedia is linked to Wikidata through a special :sameAs relationship and we are able to maximize the possibilities for semantic search results.
我们不仅可以使用可能从未用作关键词的搜索词在系统中找到图像，而且我们还发现我们可以快速发现 DBpedia 和 Wikidata 知识模型中的不一致之处。
Not only can we find images in the system using search terms that would have likely never been used as keywords, but we have also found we can rapidly spot inconsistencies in the DBpedia and Wikidata knowledge models.
当我们第一次注意到针对与搜索条件不太匹配的查询找到的图像时，我们开始深入挖掘。有时搜索结果很奇怪，因为我们根据谓词对搜索结果进行排序，但更常见的是，搜索结果很奇怪，因为 DBpedia 或 Wikidata 中 SKOS 关系或子类的构建方式。
When we first started noticing images found for queries that didn’t quite fit the search, we began digging deeper. Sometimes search results are odd because we sort our search results based on the predicate, but more often, the search results are odd because of the ways in which the SKOS relationships or subclasses have been constructed in DBpedia or Wikidata.
这项研究启发我们创建了一个“查找搜索路径”函数，当在 ImageSnippets 中的语义搜索中找到图像时，可以调用该函数。这是一个自定义手写函数，用于模拟我们的查询，并返回搜索词返回实体的路径。
This research inspired us to create a “Find Search Paths” function that can be called when an image is found in a semantic search in ImageSnippets. It is a custom handwritten function for emulating our queries and returns the path in which an entity was returned for a search term.
图 9.8 ImageSnippets 中的寻路功能展示了从包含实体“cold”的三元组到 Wikidata 中“media”概念实体的路径。
Figure 9.8 The pathfinding function in ImageSnippets shows a path from a triple with the entity “cold” linked to an entity for the concept of “media” in Wikidata.
图 9.8 显示了搜索概念“媒体”返回的图像。
Figure 9.8 shows an image that was returned for a search for the concept, “media.”
尽管“媒体”是一个高级抽象概念，但符合媒体概念的图像往往与报纸、广告、海报、书籍、杂志等主题相关。ImageSnippets 系统中目前有近 1,500 张图像用于查询“媒体”实体，并且大多数图像都很容易与该查询匹配，即使这些图像都没有标注“媒体”概念。
Even though “media” is a high level, abstract concept, images that match the media concept tend to relate to topics like newspapers, advertising, posters, books, magazines and so on. There are currently close to 1,500 images in the ImageSnippets system that are returned for a query for the “media” entity and most are easy enough to match to that query even though none of those images were annotated with the concept “media.”
在搜索“媒体”时，偶尔会有一些毫无意义的图片脱颖而出。其中一张非常容易发现的图片是一张棕榈叶上的冰冻照片。
Every now and then an image will stand out in a search for “media” that really doesn’t make sense. One of the images that was extremely easy to spot was that of a photo of frozen ice on a palm leaf.
我们的“查找搜索路径”函数返回了 DBpedia 实体“cold”与 Wikidata 概念“media”之间的以下路径。DBpedia“Cold”包含指向 Wikidata 中相同概念的 :sameAs 链接。然而，在概念“sensation”和“detection”与 Wikidata https://www.wikidata.org/wiki/Q5720030 中的实体（其中 detection 是“information access”的子类）之间发生了一些奇怪的事情。
Our “Find Search Paths” function returned the following path between the DBpedia entity “cold” to the Wikidata concept of “media.” DBpedia “Cold” contains a :sameAs link to the same concept in Wikidata. However, something odd happens between the concept “sensation” and “detection” and the entity in Wikidata https://www.wikidata.org/wiki/Q5720030 in which detection is given as a subclass of “information access.”
dbpedia:Cold 与 wikidata:Q270952 中的“Cold”相同
dbpedia:Cold is :sameAs “Cold” in wikidata:Q270952
属于“感觉”的一个子类 wikidata:Q3955369
which is a subclass of “sensation” wikidata:Q3955369
属于“检测”的子类 wikidata:Q5720030
which is a subclass of “detection” wikidata:Q5720030
属于“信息访问”的子类 wikidata:Q593289
which is a subclass of “information access” wikidata:Q593289
它是“频道”的子类 wikidata:Q733553
which is a subclass of “channel” wikidata:Q733553
属于“媒体”的子类 wikidata:Q340169
which is a subclass of “media” wikidata:Q340169
维基数据中导致这种不一致的关系可能会在本书出版时得到修复，但它说明了图像在发现不一致方面的独特能力。通常，当我们在 ImageSnippets 中发现这些情况时，我们可以将它们提交给其他生态系统，以便更了解其知识模型的人可以解决它们。
The relationships in Wikidata that lead to this inconsistency will probably be fixed by the time this book is published, but it illustrates the unique ability that images can serve in spotting inconsistencies. Typically, when we discover these cases in ImageSnippets, we can refer them to the other ecosystems so that someone far more knowledgeable about their knowledge models can resolve them.
此“查找搜索路径”功能在知识图谱世界中充当一种“可解释语义”。
This “Find Search Paths” function acts as a type of “Explainable Semantics” in the knowledge graph world.
AI 增强型 PKG 构建
AI Augmented PKG Construction
ImageSnippets 系统允许从系统中调用任意数量的 AI 模型。[16] 在本章开头，图 9.2 显示了几个计算机视觉服务的屏幕截图，这些服务返回了预测分类、对象和图像字幕的混合。在该系统中，可以将算法模型与现实世界数据进行比较和评估。
The ImageSnippets system allows any number of AI models to be called from the system. [16] At the beginning of the chapter, figure 9.2 showed a screenshot of several computer vision services which had returned a mix of predicted classifications, objects and image captioning. In the system, algorithmic models can be compared and evaluated side by side against real-world data.
机器学习围绕图像描述创造了一套新词汇。机器生成的关键词预测在人工智能术语中称为分类。边界框勾勒出图像中检测到物体或特征的区域。语义分割是一种像素级预测，可识别图像中物体或特征周围的轮廓。关键点检测技术可在图像中找到突出的空间兴趣点。这对面部识别和人物检测尤其有帮助。
Machine learning has created a new vocabulary around describing images. Machine-generated predictions of keywords are called classifications in AI terminology. Bounding Boxes outline regions in an image where objects or features have been detected. Semantic segmentation is a form of pixel-level prediction that identifies outlines around objects or features in an image. The technique of keypoint detection finds spatial points of interest in an image that stand out. This helps especially with facial recognition and people detection.
统计 AI 的整个领域在很大程度上都是使用训练数据构建的，这些数据大多已被剥离了图像周围的所有非结构化、支持性元数据（无论是否嵌入）。最初，这些内容由通常不接触任何原始元数据的人工重新标记。许多较新的机器学习方法也试图越来越多地依赖几乎不使用人工标记的训练方法。
The entire field of statistical AI has, in large part, been built using training data that has mostly been stripped of all the unstructured, supportive metadata – whether embedded or not – around images. Originally, that content was then relabeled by humans who were generally not exposed to any of the original metadata. Many newer methods of machine learning also attempt to rely more and more on training methods that use little to no human labeling.
人工智能可以根据提示生成合成图像，因为它已经接受过大量文本和图像数据的训练，但这与精确解释图像场景的内容和上下文是非常不同的任务。
AI can generate synthetic images from prompts because it has been trained on massive amounts of text and image data, but this is a very different task than precisely interpreting the contents and context of image scenes.
在许多图像上，最先进的人工智能模型通常可以返回准确但完全缺乏背景的预测。
On many images, state of the art AI models can usually return predictions that are accurate, but completely lacking in context.
然而，当用于现实世界的数据时，特别是在需要专业词汇的专家领域，结果通常不那么引人注目。训练仍然需要大量带标签的图像数据集，训练数据的变化可能会发出错误的信号，算法会从中学习。
When used on real-world data however, particularly in an expert domain requiring a specialized vocabulary, the results are usually less spectacular. Training still requires large datasets of labeled images and variations in the training data can give the wrong signals from which the algorithms learn.
例如，图 9.1 中的图像被 Visual ChatGPT [17] 描述为“一堆木地板”，图 9.5 为“轮胎的特写”。图 9.2 和图 9.5 都是“海滩上的一艘船”。
For example, the image in figure 9.1 was described by Visual ChatGPT [17] as a “pile of wood flooring boards,” and figure 9.5 as a “close up of a tire.” Both figure 9.2 and figure 9.5 were “a boat on a beach.”
问题的例子多种多样，从错误地“幻觉”到再现偏见，不一而足。其他与背景有关的问题包括时间扭曲，例如当用现代物体词汇训练的模型试图对更古老的绘画进行分类时，或者空间感知问题，例如当物体太近或太远而完全被错误识别时。
Problems can be seen in examples that range from incorrectly “hallucinating” things that are not there to reproducing bias. Other problems with context have included things like temporal distortions such as when models trained on modern object vocabularies attempt to classify much older paintings, or problems with spatial perception, such as when objects are too close or too far and completely misidentified.
大部分关于要存储在图中的图像描述的个人知识都是统计长尾中的材料，本章中使用的几乎所有图像都是对机器具有挑战性的用例的示例。
Much of the personal knowledge about image descriptions to be stored in graphs is the kind of material found in the statistical long tail and almost all the images used in this chapter have been examples of use cases that are challenging for machines.
图9.9 ImageSnippets接口中调用AI元数据的函数截图。
Figure 9.9 A screenshot of the function to call AI metadata in the ImageSnippets interface.
图 9.9 展示了一张受益于 AI 增强的图像。图像中区域显示的所有动物都由计算机视觉检测器识别。实体自动映射到 DBpedia 实体。因此，机器在不到一秒的时间内创建了大约 15 个三元组。建议的区域是由机器创建的，但人类将其视为人机交互活动。甚至有人为图像建议了人类错过的区域，例如准确识别图像背景远处的斑马，而图像创建者可能没有提到这一点。
Figure 9.9 shows an image that benefited from AI augmentation. All the animals shown in regions in the image were identified by the computer vision detectors. The entities were automatically mapped to DBpedia entities. Thus about 15 triples were created by the machine in less than a second. The suggested regions were created by the machine but approved by the human as a human-in-the-loop activity. Regions were even suggested for the image that the human had missed such as the accurate identification of a zebra far in the background of this image which the image creator may have otherwise missed mentioning.
但即使在专家领域，人工智能通常也能提供有价值的建议。当与人类一起使用时，计算机视觉可以加快识别图像中要描述的区域或对象的过程，并提出人类可能未曾考虑过的分类。相反，将人类置于将非结构化数据直接映射到机器可读字段的系统的中心，也可以在以数据为中心的人工智能策略中直接反馈修改后的数据集，这是 ImageSnippets 积极开发的一个领域。
But AI can often provide valuable suggestions even in expert domains. When used with a human in the loop, computer vision can speed up the process of identifying regions or objects in images to be described and suggest classifications that a human might not have considered. Conversely, placing the human at the center of a system that maps unstructured data directly to machine-readable fields, can also directly feedback modified datasets in data-centric AI strategies and is an area of active development with ImageSnippets.
结论
Conclusion
图像在个人知识图谱中的作用是一个深刻而重要的话题。在本章中，我们探讨了图像如何在个人知识图谱中扮演辅助角色，或如何成为更复杂的图结构的基础。当图像成为知识图谱构建的核心时，它们可以激发概念之间的远距离联系并阐明知识空白。它们还可以用作本体工程师从个人知识中提取领域专业知识的过程的一部分。
The role of images in relation to personal knowledge graphs is a deep and important topic. In this chapter, we have explored how images can either play ancillary roles in personal knowledge graphs or form the basis of much more complex graph structures. When images are central to knowledge graph construction, they can stimulate distant connections between concepts and illuminate gaps in knowledge. They can also be used as part of a process for ontology engineers to extract domain expertise from personal knowledge.
为了支持这一讨论，详细描述了图像元数据。嵌入式元数据是一种非常有用但经常被忽视的商品。嵌入式元数据将不可避免地引发对位置数据隐私的考虑。然而，通过更周到地考虑嵌入式元数据，应用程序可以从允许随图像一起传输的数据中提取知识。元数据不仅可以传达用户的版权意图和描述性数据，还可以为盲人、视力低下或认知障碍用户提供一致的可访问性描述。应用程序通常对嵌入式元数据采取“全有或全无”的立场。这是一个复杂的话题，但当它包含在应用程序工作流程中时，它可以提供很多价值。
To support this discussion, image metadata was described in detail. Embedded metadata is a highly useful commodity that is often overlooked. Embedded metadata will inevitably invoke considerations about the privacy of location data. With a more thoughtful consideration of embedded metadata however, applications could extract knowledge from data that is allowed to travel with the image. Not only can metadata convey user copyright intentions and descriptive data, it can provide consistent accessibility descriptions for blind, low vision or cognitively impaired users. Applications often take an “all or nothing” stance regarding embedded metadata. It is a complicated topic, but it can provide a lot of value when included in application workflows.
ImageSnippets 系统最初是作为研究工具创建的，现在包含超过 81,000 个关于图像的命名图表，并且每天都在增长。系统中的大部分图形数据来自个人图像集。ImageSnippets 创建的三重标签是任何其他图像注释系统中都可以找到的最精确的机器可读图像描述。随着时间的推移，该系统显示出最初设计时未预料到的功能。其中一项发现是，该系统可用于识别和解释公共数据集中传递路径之间的推理不一致。本章之外可以探索的其他功能包括如何使用 ImageSnippets 图从动态 SPARQL 查询填充网站。
The ImageSnippets system, while created originally as a research tool, now contains over 81,000 named graphs about images and continues to grow daily. Much of the graphed data in the system is from personal image collections. The triple-tags created by ImageSnippets are the most precise machine-readable image descriptions that can be found in any other image annotation system. Over time the system revealed functionality not originally anticipated by the design. One of these findings was that the system could be used to identify and explain inference inconsistencies across transitive paths in public datasets. Other functionality which could be explored beyond this chapter include how the ImageSnippets graph is used to populate websites from dynamic SPARQL queries.
尽管人工智能在当代计算中具有重要意义，但本章仅对其进行了简单讨论。这是因为个人知识图谱在能够捕获机器学习技术目前无法很好解释的异常知识类型时最有用。虽然机器解释变得越来越智能，但它仍然缺乏领域知识和背景。虽然计算机视觉可以增强人类知识图谱的构建，但它也可以从学习 ImageSnippets 可以创建的结构化数据集中受益。机器可以为 ImageSnippets 系统中的知识图谱构建做出贡献的一个领域是构建三重标签所需的时间。
Artificial intelligence was only lightly discussed in the chapter despite its obvious importance in contemporary computing. This is because personal knowledge graphs are most useful when they can capture the type of outlier knowledge that is currently not well interpreted by machine learning techniques. While machine interpretation keeps getting smarter, it is still lacking in domain knowledge and context. While computer vision can augment human knowledge graph construction, it could also benefit from learning from the kind of structured datasets ImageSnippets can create. One area where machines can contribute to knowledge graph construction in the ImageSnippets system is in the time required to construct the triple-tags.
一般来说，创建元数据所需的时间与从中获得的价值始终是一个需要考虑的重要指标。不了解 RDF 的新手可以很快学会在 ImageSnippets 中构建三重标签，但随着注释者变得更加熟练，三重标签的质量会大大提高。对于许多图像，创建传统关键字和三重标签之间的时间差异可以忽略不计。然而，处理来自专家领域的图像通常非常复杂且耗时。
In general, the time it takes to create metadata versus the value that can be gained from it is always an important metric to consider. Novice users with no knowledge of RDF can be taught to build triple-tags in ImageSnippets quite quickly, but the quality of the triple-tags are much improved as annotators become more skilled. With many images, the time difference between creating traditional keywords versus triple-tags is negligible. Working with images from expert domains however can often be quite complex and time consuming.
当针对非常简单且过度使用的关键词组合返回数千张或更多图像作为语义搜索结果时，三重标签最适合于检索。例如：树、鸟、花、火车、水等。在这些情况下，即使只使用几个三重标签，也有明显的好处。
Triple-tags work best for retrieval when several thousand images or more can be returned as semantic search results for combinations of very simple overused keywords. Examples are: tree, bird, flower, train, water, etc. In these cases, even when only a few triple-tags are used, there are obvious benefits.
三重标签在含义和语境上是无歧义的，
The fact that the triple-tags are disambiguated for meaning and context, and
它们现在成为与更多概念相连的图的一部分，并可以利用链接数据推理。即使关于每张图片的少量三元组也能对个人知识图产生影响。
That they are now part of a graph connected to many more concept and can take advantage of linked data inference. Even a small number of triples about each image can make an impact in a personal knowledge graph.
我们的章节以卢浮宫之旅和我与蒙娜丽莎的自拍开始。这张图片现在既存在于我的 ImageSnippets 图表中，也组织在 Google Photos 的云端。借助链接数据注释，该图片现在直接连接到相关图片，这是一个关于达芬奇的连接资源网络，也可以通过许多其他路径找到，例如“列奥纳多·达芬奇的画作”，并能够按 LIO 谓词排序。其他链接数据系统，如 Linked Data Hub（连接到 Martynas 的作品）可以重复使用这张图片。在 Google Photos 中相对容易找到吗？是的。但我认为这样的图像非常值得为其创建图形描述吗？毫无疑问，是的。
Our chapter began with a journey to the Louvre and a selfie of me with the Mona Lisa. This image now exists both in my ImageSnippets graph as well as organized in the cloud in Google Photos. With linked data annotations, the image is now connected directly to related images, a web of connected resources about da Vinci and is also findable through many other paths such as “Paintings by Leonardo da Vinci” and able to be sorted by the LIO predicate. Other linked data systems such as the Linked Data Hub (connecting to Martynas’s work here) can reuse this image. Was it relatively easy to find in Google Photos? Yes. But do I think an image like this was very much worth creating graphed descriptions for? Unequivocally, yes.
笔记
Notes
[1] https://en.wikipedia.org/wiki/Personal_knowledge_management
[1] https://en.wikipedia.org/wiki/Personal_knowledge_management
[2] https://www.go-fair.org/fair-principles/
[2] https://www.go-fair.org/fair-principles/
[3] https://www.w3.org/TR/2014/NOTE-rdf11-primer-20140624/
[3] https://www.w3.org/TR/2014/NOTE-rdf11-primer-20140624/
[4] 概念图是组织和映射知识的图形工具（“什么是概念图”Cañas & Novak，2009。）
[4] Concept maps are graphical tools for organizing and mapping knowledge (“What is a Concept Map” Cañas & Novak, 2009.)
[5] 这项工作是一篇论文的主题：“形式化非正式，概念图和语义网的融合”（Eskridge 等，2006 年）
[5] This work was the topic of a paper: “Formalizing the Informal, A Confluence of Concept Mapping and the Semantic Web” (Eskridge, et. al, 2006.)
[6] https://iiif.io/
[6] https://iiif.io/
[7] EXIF 数据（可交换图像文件格式）自 1995 年以来就嵌入到数字图像文件中（https://en.wikipedia.org/wiki/Exif）。
[7] EXIF data, (Exchangeable Image File Format) has been embedded into digital image files since 1995 (https://en.wikipedia.org/wiki/Exif.)
[8] 自 1965 年以来，IPTC 一直在为图像元数据的传输制定标准。他们的标准最初涵盖了通过美联社和路透社等通讯社发送的新闻图像的图像描述方式。在 1990 年代，IPTC 将这些标准带到了网络上 https://photometadata.org/META-Resources-Metadata-History-Timeline。)
[8] The IPTC has been setting standards for the transmission of image metadata since 1965.Their standards originally covered the way image description accompanied news images sent over wire services such as the Associated Press and Reuters. In the 1990’s, the IPTC brought those standards to the web https://photometadata.org/META-Resources-Metadata-History-Timeline.)
[9] 该规范可以在这里找到：https://iptc.org/std/photometadata/specification/IPTC-PhotoMetadata
[9] The specification can be found here:https://iptc.org/std/photometadata/specification/IPTC-PhotoMetadata
[10] 资源描述的都柏林核心模式（https://www.dublincore.org/）起源于 20 世纪 90 年代的图书馆学。
[10] The Dublin Core schema for resource description (https://www.dublincore.org/) originated from library sciences in the 1990s.
[11] https://en.wikipedia.org/wiki/FOAF
[11] https://en.wikipedia.org/wiki/FOAF
[12] Warren, Hayes 在“Bounding Ambiguity”一文中（https://ceur-ws.org/Vol-2276/paper5.pdf）详细描述了 LIO 的构造。
[12] “Bounding Ambiguity’ Warren, Hayes (https://ceur-ws.org/Vol-2276/paper5.pdf ) describes the construction of LIO in detail.
[13] https://w3id.org/lio/v1#
[13] https://w3id.org/lio/v1#
[14] https://www.clarifai.com/
[14] https://www.clarifai.com/
[15] https://azure.microsoft.com/
[15] https://azure.microsoft.com/
[16] 知识工程与真实世界图像数据报告 (Warren, Shamma, Hayes AAAI-MAKE, 2021)
[16] Reported on in Knowledge Engineering and Real World Image Data (Warren, Shamma, Hayes AAAI-MAKE, 2021)
[17] https://huggingface.co/spaces/microsoft/visual_chatgpt
[17] https://huggingface.co/spaces/microsoft/visual_chatgpt
第十章
Chapter 10
Agora 是一个社交知识图谱
The Agora is a Social Knowledge Graph
爱德华多·伊万内茨 (弗朗西安)
介绍
Introduction
在本章中，我 [1] 描述了 Agora，这是一个具有图形、平台和协议组件的分布式系统，可帮助实践社区引导亲社会的知识共享。
In this chapter I [1] describe the Agora, a distributed system with Graph, Platform and Protocol components enabling a Community of Practice to bootstrap a pro-social Knowledge Commons.
该系统的第一个组件是 Agora Graph，这是一个分布式知识图谱 [2]，其中的节点集成自众包存储库，链接明确包含在贡献的资源中或从上下文中推断出来。[3] 将节点和链接合成到这个共享图中是第二个组件 Agora 平台的任务，该平台是一款管理贡献资源的提取和呈现的软件。前两个组件实现的常识形成过程可以通过使用可选的 Agora 协议来补充，该协议大致定义为一组惯例，由着手建立和维护 Agora 的社区出于善意应用于共同目标。
The first component of this system is the Agora Graph, which is a Distributed Knowledge Graph [2] with nodes integrated from crowdsourced repositories, and links included explicitly in contributed resources or inferred from context. [3] The synthesis of nodes and links into this shared graph is the task of the second component, the Agora Platform, which is software that manages the ingestion and presentation of contributed resources. The process of common sense-making enabled by the first two components can be supplemented with the use of the optional Agora Protocol, loosely defined as a set of conventions applied with good intent towards a shared purpose by the community that sets out to build and maintain an Agora.
请注意，本章最具体、最广泛描述的具体 Agora Commons 只是一个 Agora，它旨在作为本章探讨的愿景和原则的直接实例进行评估。截至 2023 年初，此参考实现可以简洁地描述为数字花园、wiki 和流的聚合器，这些聚合器是一组共享空间和主题，在 Agora Graph 中的对象后简称为节点。为了避免歧义，我将经常将这个原型称为 Flancia 的 Agora。[4] 这个 Agora 除了展示 Agora 系统的基本品质外，还试图代表参与社区（以下称为“Commoners”或“Agorans”）坚持强大的数字自决原则，并促进合作开展亲社会项目，重点是建设更好的 Agora。
Note that the concrete Agora Commons as described most concretely and extensively in this chapter is just an Agora, and it is meant to be evaluated as a straightforward instantiation of the vision and principles explored in this chapter. As of the beginning of 2023 this reference implementation can be succinctly described as an aggregator of digital gardens, wikis and streams into a set of shared spaces and topics, referred to simply as nodes after the objects in the Agora Graph. To prevent ambiguity, I will refer often to this prototype by name as the Agora of Flancia. [4] This Agora in particular, beyond showcasing the basic qualities of an Agora system, tries to uphold strong digital self-determination principles on behalf of the participating community (hereafter the “Commoners” or “Agorans”) and promote cooperation towards pro-social projects, with a focus on building better Agoras.
关于本章
About this chapter
如上所述，更广泛的 Agora 是一个具有多个方面的通用系统，我将按以下顺序介绍。
As hinted above, the wider Agora is a general-purpose system with several facets, which I will cover in the following order.
首先，我将介绍 Agora Graph，这是一个由社区使用本章中展示的工具和惯例提供的超图。提供的（参考）Flancia Agora 图表包含有关 Agorans 为公众利益开展的各种共享项目的信息，其中许多项目与 Agora 项目的发展直接相关。
First, I will present the Agora Graph, which is a Hypergraph provisioned by a community using tools and conventions like those showcased in this chapter. The graph of the provided (reference) Agora of Flancia contains information about a variety of shared projects being undertaken by Agorans for the public good, many of which are directly related to the development of the Agora as a project.
其次，我将介绍 Agora 平台的作用 [5]，它是一组软件包，可简化在 Agora Graph 之上提供一般有用的服务，并维护 Agora Commons 与更广泛的同意互联网之间的道德桥梁（持续的导入/导出流程）。这首先侧重于知识管理和自由软件社区的需求。
Second, I will cover the role of the Agora Platform, [5] which is a collection of software packages that simplify provisioning generally useful services on top of an Agora Graph and maintaining ethical bridges (ongoing import/export flows) between an Agora Commons and the wider consenting internet. This is done with an initial focus on the needs of the Knowledge Management and Free Software communities in particular.
第三，我将介绍 Agora 协议，我将其称为旨在降低亲社会信息流和合作阻力的任何协议。在本章中，我提出了一个简单的默认方案，该方案试图建立在知识导向实践社区中已经使用的惯例和设备之上，就像在撰写本文时围绕各种思维工具出现的惯例和设备一样。
Third, I will present a sketch of Agora Protocol, which is what I call any protocol designed to lower friction to pro-social information flow and cooperation. In this chapter I propose a simple default that tries to build on conventions and devices already in use within knowledge-oriented Communities of Practice like those emerging around a variety of Tools for Thought as of the time of writing.
第四，也是最后一点，我将在“弗朗西亚集市的目标之旅”部分探讨整合提议系统的各个方面可能产生的应用。这包括讨论如果至少一个广泛可用的知识共享空间得以实现（即成为社区的焦点）可能产生的结果。
Fourth, and finally, I will explore possible applications arising from the integration of all facets of the proposed system in the section A tour of the goals of the Agora of Flancia. This includes some discussion of potential outcomes if at least one widely available Knowledge Commons is brought to fruition, in the sense of becoming a Focal Point for a community.
定义
Definitions
该项目范围广泛，使用来自不同领域的术语，从计算机科学、系统思维、政治理论和数学中汲取灵感。它还利用隐喻试图向一些首次阅读本文的读者阐明本文描述的潜在认知设备的实用性。为了帮助理解，我在此提供了要遵循的关键术语的简短摘要。您可以随意收藏此部分，跳过它并在需要时返回它。
This project has a large scope and makes use of terminology from different fields, drawing inspiration from computer science, systems thinking, political theory, and mathematics. It also makes use of metaphors to try to illuminate the usefulness of potential cognitive devices described here to some readers for the first time. To aid understanding, here I provide a short summary of key terms to follow. Feel free to bookmark this section, skip it and return to it as the need arises.
Agora：具有协议、平台和图形组件的知识共享空间。互联网的集成子集，正确理解：用作集体思考的工具。
The Agora: A Knowledge Commons with Protocol, Platform, and Graph components. An integrated subset of the internet, held right: used as a tool for collective thought.
Flancia 的 Agora：参考或示例 Agora 实例，即截至撰写本文时可在 https://anagora.org 和 https://flancia.agor.ai [6] 上浏览的实例。Flancia 的 Agora 专注于数据交换作为互操作性的核心机制。它可以从 git 存储库导入常见格式的资源并提供相同的集成视图。
The Agora of Flancia: A reference or example Agora instance, meaning the one browsable at https://anagora.org and https://flancia.agor.ai [6] as of the time of writing. The Agora of Flancia focuses on data exchange as the core mechanism for interoperability. It can import resources in common formats from git repositories and offer integrated views of the same.
Agora Graph：Agora 的大脑，具体指由社区维护的连接组。由用户创建、标记和注释的资源集成而成的分布式社交 [7] 知识图谱。
Agora Graph: The brain of the Agora, meaning specifically a connectome maintained by a community. A distributed social [7] knowledge graph integrated out of resources created, tagged and annotated by users.
Agora 平台：Agora 的核心，即有用的泵，具体来说是一套帮助共享的服务。截至撰写本文时，该平台的用户可见方面包括 Web 应用程序、API 以及一组用于各种数据源和格式的导入器和导出器 - 全部为免费软件。
Agora Platform: The heart of an Agora, in the sense of a useful pump, concretely a set of services aiding Commoning. The user visible aspects of this platform as of the time of writing include a web application, an API, and a set of importers and exporters for diverse data sources and formats – all free software.
Agora 协议：Agora 的语言。社区使用该协议可让普通人清晰地沟通，并让 Agora 实例更好地进行语义解析，并将用户资源整合到更大的 Agora 网络中联合的断言和亲社会意图列表中。
Agora Protocol: The language of an Agora. Its usage by a community enables Commoners to communicate clearly and Agora instances to better semantically parse and integrate user resources into lists of assertions and pro social intents federated within the greater Agora Network.
Agora 编辑器：允许用户向 Agora 写入内容的各种工具之一。这可以直接进行（通过写入 Commons 中的存储库）或间接进行（通过桥接或 Agora 机器人进行交叉发布）。
Agora Editor: Any of a variety of tools that allows a user to write to an Agora. This can happen either directly (by writing to a repository in the Commons) or indirectly (via cross posting through a bridge or Agora bot).
节点：Agora Graph 中的顶点。映射到从资源名称或附加元数据解码的主题或实体描述。可以通过 wikilink 或 #hashtag 引用。子节点的集合。
Node: A vertex in the Agora Graph. Maps to a topic or entity description decoded from a resource’s name or attached metadata. Can be referred to by wikilink or #hashtag. A collection of subnodes.
子节点：用户贡献的与节点上下文相关的资源。默认情况下，当查询节点时，Agora 会显示按社区排名的子节点集合。
Subnode: A resource contributed by a user as relevant in the context of a node. An Agora will surface a community-ranked collection of subnodes when queried for a node by default.
链接：Agora Graph 中节点之间的注释边。社区贡献的概念或上下文之间的关系或联系。[8]
Link: An annotated edge between nodes in the Agora Graph. A relationship or connection between concepts or contexts as contributed by the community. [8]
前向链接：或传出链接。如果 A 链接到 B，则我们称 A 中存在到 B 的前向链接。也称为传出链接。
Forward Link: Or outgoing link. If A links to B, then we say in A there is a forward-link to B. Also known as outgoing link.
反向链接：如果 A 链接到 B，那么我们说 B 中有一个指向 A 的反向链接。也称为入站链接。
Back Link: If A links to B, then we say in B there is a back link to A. Also known as incoming link.
数字花园：由个人长期维护的相互关联的资源库。用户向 Agora Graph 贡献的个人知识图谱，其中大多数贡献的时间顺序并不像其累积效应和连接性那么重要。
Digital Garden: A repository of interlinked resources maintained by a person over time. A Personal Knowledge Graph contributed by a user to the Agora Graph in which the chronological aspect of most contributions is not as relevant as their cumulative effect and their connectivity.
流：时间轴中表示一系列事件的资源集合。社交网络信息流和博客或数字花园中的一系列日记条目都可以视为流。
Stream: A collection of resources in a timeline representing a sequence of events. A social network feed and a series of journal entries in a blog or digital garden can all be seen as streams.
围墙花园：一个封闭的社交平台，特别是指随着时间的推移和为私人利益而阻碍信息流动或以其他方式对内容和访问方式进行控制的平台。
Walled garden: A closed social platform, in particular one that hinders the flow of information or otherwise exerts control over content and means of access, over time and for private benefit.
公地：一种“超越市场和国家”的实体（Ostrom，2009）。由社区为某一目的提供和维护的活系统。[9] 公地及其朋友可获得的资源和服务的集合。
Commons: An entity “beyond markets and states” (Ostrom, 2009). A living system provisioned and maintained by a community for a purpose. [9] The set of resources and services available to Commoners and their friends.
桥梁：连接环境并实现资源和信息流动的设备或流程。可用于提供围墙花园和公共空间等之间的互操作性。
Bridge: A device or process connecting contexts and enabling the flow of resources and information. Can be used to provide interoperability between e.g., walled gardens and the Commons.
影响
Influences
Agora 是一个范围广泛的项目，因此它建立在许多人的工作之上。在本节中，我将提到该项目的一些核心影响，并尝试提出一个统一的愿景，即 Agora 是一个用于大规模整合和合作的通用工具。
The Agora is a project with a broad scope and accordingly builds on the work of many. In this section I mention some of the core influences of the project and try to present a cohesive vision of the Agora as a general-purpose vehicle for integration and cooperation at scale.
集市可以最广泛地定义为互联网的一个共享子集，用于建设性和特定目的。从这个意义上说，弗兰西亚的集市受到了埃莉诺·奥斯特罗姆 (Elinor Ostrom) 作品的启发，她率先研究了世界各地运作良好的公地。[10] 她的工作消除了曾经占主导地位的立场，即公地必须由国家和公司等集中代理机构接管 (封闭)，以防止所谓的公地悲剧。在她职业生涯的后期，她在同样具有影响力的作品中特别探索了知识公地。[11]
An Agora can be defined most generally as a shared subset of the internet used constructively and towards a particular purpose. In this sense the Agora of Flancia is inspired by the work of Elinor Ostrom, who pioneered the study of well-functioning Commons around the world. [10] Her work resulted in dispelling the once dominant position that the Commons must invariably be taken over (enclosed) by centralized agents like the state and corporations to prevent the so-called Tragedy of the Commons. Later in her career she explored the Knowledge Commons in particular in work which is equally influential. [11]
稍微不那么笼统地说，如上一节所述，Agora 可看作是一个平台和协议，用于配置和维护一个图谱，使社区能够明确定义和推进他们的目标。这意味着它是一种用于共同回答问题和解决问题的工具，从自治和政策制定开始，但推广到 Agorans 感兴趣的所有问题。Agora 的这个方面（和潜力）很大程度上归功于 Christopher Alexander 的工作，他提出了模式的概念，即抽象（但可实例化）的问题解决设备或模块，可以由社区实现。[12] Flancia 的 Agora 尤其试图看起来和感觉起来像一个模式存储库，这意味着它试图集成和扩展各种互补的模式语言。
Slightly less generally, and as per the previous section, an Agora can be seen as a Platform and a Protocol for provisioning and maintaining a Graph enabling a community to clearly define and advance their goals. This means it is a tool meant for answering questions and solving problems together, starting with those of self-governance and policy making but generalizing to all problems that the Agorans find interesting. This aspect (and potential) of the Agora owes heavily to the work of Christopher Alexander, who developed the notion of Patterns as abstract (but instantiable) problem solving devices or modules that might be implemented by a community. [12] The Agora of Flancia in particular tries to look and feel like a Pattern Repository, meaning it tries to integrate and extend a variety of complementary Pattern Languages.
Silke Helfrich 和 David Bollier 等人明确地整合了前两门学科（共享和模式化），并积极致力于为公共领域制定明确的模式语言[13]，作为社会改革的力量，而 Flancia 的 Agora 试图使用、扩展和推广这种模式语言。
Silke Helfrich and David Bollier among others explicitly integrated the previous two disciplines (Commoning and Patterning) and worked actively on an explicit Pattern Language for the Commons [13] as a force for social reform, which the Agora of Flancia tries to use, extend and promote.
Vannevar Bush、Douglas Engelbart、Alan Kay 和 Ted Nelson 是“思维工具”领域的一些远见卓识者，他们几乎在该领域的每一篇论文中都得到了感谢（有理由的），并为 Flancia 的 Agora 背后的一些技术设计决策提供了信息。参考 Agora 实现于 2020 年开始，其直接目标是构建一个社交 Memex，它大量使用了 Transclusion 作为设备。后者归功于 Ted Nelson，模式 Intertwingularity 也是如此：
Vannevar Bush, Douglas Engelbart, Alan Kay, and Ted Nelson are some of the visionaries in the field of Tools for Thought who are thanked in almost every paper in this space (with reason) and who inform some of the technical design decisions behind the Agora of Flancia. The reference Agora implementation was started in 2020 with the direct goal of building a Social Memex and it makes heavy use of Transclusion as a device. The latter is owed to Ted Nelson, as is the pattern Intertwingularity:
等级和顺序结构，尤其是自古腾堡以来流行的结构，通常是强制的和人为的。互文奇点通常不被承认——人们一直假装他们可以让事物等级化、可分类和顺序化，但实际上他们做不到。
Hierarchical and sequential structures, especially popular since Gutenberg, are usually forced and artificial. Intertwingularity is not generally acknowledged – people keep pretending they can make things hierarchical, categorizable and sequential when they can’t.
（纳尔逊，1974 年）
(Nelson, 1974)
总而言之，可以说，Agora 是对 Intertwingularity 和 Heterarchy 等模式的潜力的一次公共探索，这些模式可以扩大规模、进行组合，并在公共空间中产生有趣的突发行为。[14]
All in all, it could be said that the Agora is a public exploration of the potential of patterns like Intertwingularity and Heterarchy as they scale up, compose, and yield interesting emergent behavior in the Commons. [14]
在本世纪初，我热衷于阅读 Everything2 (E2)，这可能在我知道存在这种模式之前就形成了 Agora 松散、机会主义合作的基本模式。该网站的内容（其中大部分现已删除）范围从百科全书（维基百科之前）到创意。E2 社区接受了点名作为一种活动的概念，这意味着在标记（语义映射）位置贡献独立写作。
I was an avid reader of Everything2 (E2) around the turn of the century and that likely shaped the basic Agora patterns of loose, opportunistic collaboration since before I knew that patterns existed. The content of the site (much of it now taken down) ranges from the encyclopedic (before Wikipedia) to the creative. The E2 community embraced the concept of noding as an activity, meaning contributing independent writeups in labeled (semantically mapped) locations.
在技术基础层面，整个互联网都归功于 Tim Berners-Lee，他首先在 Web 上的工作，现在又在 Solid 上的工作，这为 Agora 提供了 Flancia 对数字自决的方法，并且可能是我们将来构建更好的 Agora 的平台之一。
On the technology ground level the whole internet is indebted to Tim Berners-Lee for his work first on the Web and currently on Solid, which informs the Agora of Flancia’s approach to digital self-determination and is likely one of the platforms on which we will build better Agoras in the future.
沃德·坎宁安（Ward Cunningham）开发了第一个 Wiki 和原始的 c2 模式存储库，从而彻底改变了知识空间，他继续通过 FedWiki 激励我们，FedWiki 探索了与 Agora 兼容的分布式协作模式。[15]
Ward Cunningham shaped the knowledge space for good by developing the first Wiki and the original c2 Pattern Repository, and he continues to inspire us with FedWiki, which explores patterns for distributed collaboration compatible with those of the Agora. [15]
维基社区继续在这个领域做出大量创新贡献。在最初的 WikiWikiWeb 中开发的 Wiki 模式可以看作是讨论和并发编辑的模式。Interwiki、TwinPages 和 SisterSites 是跨维基连接和机会性协作的模式。Mycorrhiza 是一个支持 Hypha 概念的维基引擎，它类似于 Agora 节点，因为它包含（或通向）多种资源。Massive Wiki 是一个基于文件和 git 的维基引擎，其互连和集成目标与 Agora 的目标兼容。
The wiki community continues to contribute a lot of innovation in this space in general. Wiki Modes as developed in the original WikiWikiWeb can be seen as patterns for discussion and concurrent editing. Interwiki, TwinPages, and SisterSites are patterns for cross-wiki connection and opportunistic collaboration. Mycorrhiza is a wiki engine supporting the concept of Hypha, which resembles an Agora node in that it contains (or leads to) a multiplicity of resources. Massive Wiki is a file and git-based wiki engine with interlinking and integration goals compatible with those of an Agora.
除了 wiki 空间之外，Codex Editor 还是一个统一的知识管理平台，其愿景是成为“知识工作者的操作系统”，该平台围绕独立属性构建，而不是内联元数据（如标签和 wikilinks）。Cosma 是一款与工具无关的可视化和共享工具，适用于知识工作者，可在 Web 界面中呈现相互链接的纯文本文件。Subsconscious 的 Noosphere 看起来可以充当高级（完全分布式）Agora。
Beyond the wiki space, Codex Editor is a unified knowledge management platform with a vision of becoming a “Knowledge Workers’s OS” built around standoff properties as opposed to inline metadata like hashtags and wikilinks. Cosma is a tool-agnostic visualization and sharing tool for knowledge workers which renders interlinked plain text files in a web interface. Noosphere by Subsconscious looks like it could function as an advanced (fully distributed) Agora.
有关社交知识空间或与 Agora 项目相关的项目的最新列表，请参阅 Flancia 的 Agora 中的节点 Agora Space。
For an up-to-date list of projects in the social knowledge space or otherwise related to the Agora project, please refer to node Agora Space in the Agora of Flancia.
在 Agora Commons
In the Agora Commons
当我说 Agora 不合格时，我指的是最终具有联邦性质的总体 Agora Commons。它首先是一个知识共享，但它渴望获得一个物理维度，让其社区能够规划、提供和管理社会项目。
When I say the Agora unqualified, I mean an overarching Agora Commons eventually of federated nature. It is a Knowledge Commons to begin with, but one aspiring to gain a physical dimension that lets its community plan, provision and manage social projects.
图 10.1 公共三要素（Bollier & Helfrich，2019）为 Agora Commons 提供了信息。[16]
Figure 10.1 The Triad of Commoning (Bollier & Helfrich, 2019) informs the Agora Commons. [16]
普通人可以选择在通过 Agora 平台向 Agora Graph 贡献资源时使用 Agora 协议。通过这种组合，社区可以轻松地众包一组有趣的资源和意图，并通过它们在 Commons 中提供有用的服务。在以下部分中，我将尝试进一步详细介绍 Agora Commons 的现状和可能情况。
Commoners can elect to use Agora Protocol in resources they contribute to an Agora Graph through an Agora Platform. The combination allows a community to easily crowdsource sets of interesting resources and intents and, through them, provision useful services in the Commons. In the following sections I try to further detail these aspects of the Agora Commons as it is and as it could be.
Agora Graph
Agora Graph
Agora Commons 首先是围绕社区选择构建的知识图谱构建的。他们将资源库和个人资源贡献给共享池，然后让 Agora 平台使用名称关联和其他简单的启发式方法将它们整合成一个社会（最终具有凝聚力）整体。[17]
An Agora Commons is built first and foremost around a Knowledge Graph that the community chooses to build. They do so by contributing repositories with resources and individual resources to a shared pool, and letting the Agora Platform integrate them using name association and other simple heuristics into a social (eventually cohesive) whole. [17]
在此图中，节点是一个集合；它是所有已知资源的集合，这些资源与节点 ID 描述的实体有关（或相关），节点 ID 仅表示其标题或规范的 wikilink（您要写入的内容以链接到它）。边是一个链接，是节点之间可选注释的关系；简而言之，如果一个节点中的资源包含指向另一个节点的 wikilink，则这两个节点在 Agora Graph 中被视为通过边连接。
In this graph, a Node is a collection; it is the set of all known resources about (or related to) the entity described by the node ID, meaning simply its title or canonical wikilink (what you would write to link to it). An Edge is a link, being an optionally annotated relationship between nodes; put simply, if a resource in one node contains a wikilink to a different node, the two nodes are considered connected by an edge in the Agora Graph.
图 10.2 由于节点 A 和节点 B 之间至少有一个子节点到子节点的链接，因此 Agora 将构建更高级别的节点到节点关系。
Figure 10.2 Because there’s at least one subnode-to-subnode link between Node A and Node B, an Agora will construct a higher level node-to-node relationship.
这里，用户 Bob 和 Zoe 可能会从 Alice 的链接中受益。
Here, users Bob and Zoe may benefit from Alice’s linking.
在 Flancia 的 Agora 中，节点之间的链接默认由上下文注释。图中的边携带有关其构建上下文的信息，因为它们本质上被认为是由相关链接周围的 #hashtags 和 wikilinks 标记的。这意味着图中实体对之间的链接是多个，并在其（唯一）标签中携带信息，可以说是注释了关系。这种多样性使 Agora Graph 有效地成为一个统一的超图，从而提供了灵活性和通用性的强大保证。[18] Flancia 的 Agora 模糊链接和节点之间区别的方式也让人想起递归超图或 Ubergraph。[19]
In the Agora of Flancia, links between nodes are annotated by context by default. Edges in the graph carry information about the context in which they were built, as they are essentially considered to be labelled by #hashtags and wikilinks found around the link in question. This means links between pairs of entities in the graph are multiple and carry information in their (unique) labels, which can be said to annotate the relationships. This multiplicity makes the Agora Graph effectively a unified hypergraph, which gives strong assurances of flexibility and generality. [18] The way the Agora of Flancia blurs the distinction between links and nodes is also reminiscent of an Recursive Hypergraph or Ubergraph. [19]
举个例子。在 Flancia 的 Agora 中，在节点 [[Agora of Flancia]] 中的资源中找到以下 Agora 协议块将导致节点 [[Agora of Flancia]] 在 Agora Graph 中通过注释管理员链接到 [[Flancia Collective]]，[20] 并分别通过注释 git 和 go 链接到 URL https://github.com/flancian/agora 和 https://anagora.org：[21]
An example might be in order. In the Agora of Flancia, finding the following Agora Protocol blocks in a resource in node [[Agora of Flancia]] will result in node [[Agora of Flancia]] being linked in the Agora Graph to [[Flancia Collective]] with the annotation steward, [20] and being linked to URLs https://github.com/flancian/agora and https://anagora.org with the annotations git and go respectively: [21]
#steward [[Flancia Collective]]
#steward [[Flancia Collective]]
#git https://github.com/flancian/agora
#git https://github.com/flancian/agora
#前往 https://anagora.org
#go https://anagora.org
由于上述区块在节点之间建立了边线，Agora 平台可以为复合查询提供附加功能，例如 [[Flancia/steward 的 Agora]]，在节点的交叉点处显示区块并拉动 [[Flancia Collective]]，以及 [[flancia 的 go/agora]] 和 [[flancia/git 的 go/agora]] 重定向到关联的 URL。有关更多信息，请参阅 [[Agora Actions]] 部分，特别是 [[Go Links]]。
Because of the edges set up between nodes by the above blocks, the Agora Platform can offer additional functionality for a compound query such as [[Agora of Flancia/steward]], presenting the blocks in the intersection of the nodes and pulling [[Flancia Collective]], and [[go/agora of flancia]] and [[go/agora of flancia/git]] redirecting to the associated URLs. See the sections on [[Agora Actions]] and in particular [[Go Links]] for more.
图 10.3 局部上下文图示例。后退和前进链接以及操作 #pull 和 #push 本质上是社交性的；许多不同的用户在 Flancia 的 Agora 中贡献了这些链接。
Figure 10.3 Example of a local context graph. Back and forward links, and actions #pull and #push, are social in nature; many different users have contributed these links in the Agora of Flancia.
排行
Ranking
Agora Graph 中的节点和链接本质上是社交性的，每个节点和链接都是一个多重性（一个集合），是整个社区为共享资源池做出贡献的结果。这引出了一个问题：应该如何处理图中的搜索查询，特别是当存在冲突或歧义时。参考 Agora 平台提供了基本的查询功能，我将在本节中简要介绍。有关平台提供的服务的更多信息以及参考实现中探讨的排名算法的详细信息，请参阅下面的相关部分。
Nodes and links in the Agora Graph are social in nature, each node and link being a multiplicity (a set) and the result of the whole community contributing into a shared resource pool. This brings up the question of how search queries into the graph should be handled, in particular when there are conflicts or ambiguities. The reference Agora Platform provides basic querying capabilities which I cover shortly in this section. For more on services provided by the platform and details about ranking algorithms explored in the reference implementation, see the relevant sections below.
当节点中的不同资源以不同的方式定义与目标节点的关系时，Flancia 的 Agora 会尝试以默认意图来处理匹配查询，即复合、集成和统一。这意味着，将向表示所有贡献（和意见多样性）的超图添加链接作为注释，并且查询将倾向于返回并集，并希望返回有用的（建设性的）组合。当无法进行合并或组合时（例如，因为查询需要返回单个结果），Agora 将应用社区和用户设置的排名规则来确定返回的内容。
When different resources in a node define a relationship to a target node differently, the Agora of Flancia tries to handle matching queries with the default intent to compound, integrate, and unify. This means that links will be added to the hypergraph representing all contributions (and diversities of opinion) as annotations, and that queries will tend to return unions and hopefully useful (constructive) compositions. When coalescing or composing is not feasible (e.g., because the query requires a single result to be returned), the Agora will apply community- and user-set ranking rules to determine what is returned.
让我们举个例子来说明这一点。假设用户 Alice 和 Bob 都为节点 [[Foo]] 做出贡献，因为他们想为他们最喜欢的 foo 提供商提供 URL。因此，Alice 贡献了一个包含块 #go https://foobar.com 的资源，而 Bob 贡献了块 #go https://foobaz.org。此后，访问 [[Foo]] 将按顺序显示这两个 URL。此外，访问 [[go/Foo]] 将触发操作 #go 的默认行为并重定向到两个 URL 中的一个，可选择在短时间内显示这两个 URL。选择哪一个（例如，较早或较晚贡献的那个，或者属于最老用户的那个）取决于社区设定的规则。Flancia 的 Agora 目前实际上是随机选择的，直到其社区定义了更好的政策。
Let’s take an example to illustrate this. Say users Alice and Bob both contribute to node [[Foo]] because they want to volunteer URLs for their favorite foo providers. So, Alice contributes a resource which includes the block #go https://foobar.com and Bob contributes the block #go https://foobaz.org. After this, visiting [[Foo]] surfaces both URLs sequentially. Additionally, visiting [[go/Foo]] will trigger the default behavior for action #go and redirect to one of the two URLs, optionally after showing both for a short period of time. Which one is picked (e.g., the one contributed earlier or later, or the one belonging to the oldest user) depends on community-set rules. The Agora of Flancia currently effectively chooses at random until a better policy is defined by its community.
请注意，Agora Graph 的作者和注释者，以及更广泛意义上的 Commons 中的活跃参与者，都承担着一种责任，即他们的贡献和行动会对读者和其他作者产生影响。要举例说明这种影响，请考虑 Agora 试图就实体的规范定义达成一致，或定义要为 [[Agora Search]] 中的查询返回的有用 URL 的最佳排名列表的问题。[22] 为了帮助他们安全有效地承担这一责任，Commoners 被提供 Agora 平台。
Note that writers and annotators of the Agora Graph, and more generally the active participants in the Commons, carry a responsibility in the sense that their contributions and actions have an impact on readers and other writers. For an example of this impact, consider the problem of an Agora trying to agree on a canonical definition for an entity, or defining the optimal ranked list of useful URLs to be returned for a query in [[Agora Search]]. [22] It is to help them carry this responsibility safely and usefully to cooperative fruition that Commoners are provided with an Agora Platform.
Agora 平台
Agora Platform
这里我们介绍了现有 Agora 平台的一些方面，该平台支持各种现成的思考工具作为 Agora 编辑器，并尝试为社区提供普遍有用的网络服务。
Here we cover some aspects of the existing Agora Platform which supports a variety of off-the-shelf Tools for Thought as Agora Editors and tries to provision generally useful web services for the community.
这是 Agora 系统最扎根于软件领域的方面，因此它在很大程度上受到现有参考实现的影响。因此，在本节中，我将具体关注 Flancia 的 Agora 提供的当前和计划中的用户旅程，深入介绍用户体验和用户界面方面，并特别强调对个人和社会知识管理社区具有潜在价值的方面。
This is the facet of the Agora system that is most anchored to the realm of software, and consequently it’s heavily shaped by the existing reference implementation. Therefore, in this section I focus concretely on the current and planned user journeys provided by the Agora of Flancia, covering user experience and user interface aspects in some depth, and stressing in particular those of potential value to the Personal and Social Knowledge Management communities.
編輯
Editors
任何可用于维护存储库并随后将其贡献给 Agora 的程序都可以充当有用的 Agora 编辑器，无论它是否具有任何明确允许或进一步集成。参考 Agora 的核心设计原则之一是尝试保持与工具无关，因此 Flancia 的 Agora 是存储库驱动的。
Any program that can be used to maintain a repository that is then contributed to an Agora can function as a useful Agora Editor regardless of whether it counts with any explicit allowances or further integrations. One of the core design principles of the reference Agora is to try to remain tool agnostic, and because of this the Agora of Flancia is repository driven.
截至撰写本文时，人们正在向 Agora 贡献使用各种工具和格式维护的存储库：Logseq、Obsidian、Foam、TiddlyWiki、Org Mode、Federated Wiki、Mycorrhiza、Wiki Vim 和 Roam Research 都是我们的用户群。一些用户经常同时使用多种工具编辑他们的存储库。[23]
As of the time of writing, people are contributing repositories maintained with a variety of tools and formats to the Agora: Logseq, Obsidian, Foam, TiddlyWiki, Org Mode, Federated Wiki, Mycorrhiza, Wiki Vim, and Roam Research are all represented in our user base. Some users routinely edit their repositories with multiple tools concurrently. [23]
当编辑花园或存储库并希望为 Agora 做出贡献时，我们鼓励使用乐观链接。这是因为 Agora 中的其他人可以“填补”给定个人知识图谱中尚未指向任何地方的链接。[24]
When editing a garden or repository with the intent to contribute to an Agora, Optimistic Linking is encouraged. This is because links that don’t (yet) lead anywhere in a given Personal Knowledge Graph can be “filled in” by others in an Agora. [24]
除了整个存储库的贡献之外，Flancia 的 Agora 还支持个人资源贡献。这是社交媒体（Fediverse 和 Twitter）和聊天（Matrix）等第三方平台上可用的默认贡献模式，其工作方式如下。
Beyond whole-repository contributions, the Agora of Flancia also supports individual resource contributions. This is the default mode of contribution available on third party platforms such as social media (the Fediverse and Twitter) and chat (Matrix), and works as follows.
首先，用户可选择添加或关注 Agora 机器人帐户。如果他们跳过此步骤，其余步骤仍可执行，但要求在所有贡献中明确提及 Agora 机器人帐户。
First, optionally, the user needs to add or follow an Agora bot account. If they skip this step, the remainder of the procedure will still work but will require that the Agora bot account is mentioned explicitly on all contributions.
其次，用户必须在帖子中使用至少一个 [[wikilink]] 或 #hashtag 来表示将其链接到匹配节点的意图。这将导致 Agora 机器人帐户使用所提及节点的 URL 回复。
Second, the user must use at least one [[wikilink]] or #hashtag in a post to signal an intent to link it in the matching nodes. This will cause the Agora bot account to reply with URLs to the nodes that were mentioned.
截至撰写本文时，Flancia 的 Agora 默认链接但不保存社交媒体贡献。用户可以选择自动保存资源，这意味着将资源存储在用户（首选）或 Agora（尚待进一步设置）控制的存储库中。他们还可以选择让 Agora 机器人对其他约定做出反应（以这种方式扩展 Agora 协议）。目前，选择加入是通过向特定节点做出贡献来标记的，这些节点表示同意相关操作行为。有关当前实施的详细信息，请参阅 Agora Bridge 部分。
As of the time of writing, the Agora of Flancia links, but doesn’t save, social media contributions by default. Users can opt into automatic saving of resources, meaning storage of resources in a repository under the users’ (preferred) or an Agora’s (while further setup is pending) control. They can also opt into Agora bot reacting to additional conventions (in this way extending Agora Protocol). Opt-in is currently flagged by contributing to specific nodes that signal consent to associated action behavior. For details on the current implementation, please see the section on Agora Bridge.
用户界面
User Interface
在下一部分中，我将介绍参考 Agora 平台为用户提供的一些 Agora Graph 查询和渲染服务。
In this next section I cover some of the Agora Graph querying and rendering services the reference Agora Platform provides to users.
从高层次来看，用户界面为以下用户旅程提供了视图和其他功能：
At a high level, the user interface provides views and other affordances for the following user journeys:
（a）查询 Agora Graph 中的节点和关系；
(a) querying nodes and relationships in the Agora Graph;
（b）可视化丰富的内容；
(b) visualizing rich contexts;
（c）执行 Agora 行动；
(c) executing Agora Actions;
(d) 在 Agora 和网络其他地方发现相关资源，用于研究或娱乐。
(d) discovering relevant resources, for research or entertainment, in-Agora and elsewhere on the web.
Flancia 的 Agora 的主要视图是节点视图。其主要设计愿景是成为一个最小可行平台，使合唱团能够汇聚到新兴意义。它试图为读者和作者提供有用的集成上下文，用于（或关于）范围内的实体。默认情况下，Agora 节点显示为与节点（子节点）关联的 Agora 内资源序列；后面是拉入其中的任何节点或资源；后面是 Stoas（稍后会介绍更多）；后面是 Agora 服务器乐观地自愿提供的其他（可能相关的）节点。在此默认视图的顶部，Agora 提供指向网络搜索的直接链接，并显示匹配的维基百科页面和 Wikidata 实体（如果已知）。
The main view of the Agora of Flancia is the node view. Its main design vision is to be a minimal viable platform enabling a chorus of voices to converge on emergent meaning. It tries to provide to both readers and writers a useful integrated context for (or about) the entity or entities in scope. An Agora node, by default, is presented as a sequence of in-Agora resources associated with the node (subnodes); followed by any nodes or resources pulled in those; followed by Stoas (of which more in a minute); followed by other (maybe related) nodes volunteered optimistically by the Agora Server. At the top of this default view, the Agora provides direct links to web searches, and surfaces matching Wikipedia pages and Wikidata entities if known.
图 10.4 Agora 节点视图示例，整合了不同用户贡献的相关文本和媒体（https://anagora.org/flaneur）。
Figure 10.4 Example Agora node view integrating relevant text and media contributed by different users ( https://anagora.org/flaneur ).
与节点关联的柱廊通常为柱廊。柱廊是一种特殊的资源，充当嵌入在 Agora 中的半公共空间，旨在支持协作和促进合作。
Associated with a node you will often find Stoas. Stoas are a special kind of resource which behaves as a semi-public space embedded in the Agora with the intention to support collaboration and foster cooperation.
可以通过将资源标记为 Stoa 来将其声明为 Stoa。[25] 在撰写本文时，将资源声明为 Stoa 的主要结果是 Agora 默认将该资源视为可嵌入，并尝试在相关上下文中提供它。Flancia 的 Agora 还尝试在每个节点中提供有用的默认 Stoas：至少一个公共（当前可由世界写入）文本文档，允许并发实时编辑，[26] 和一个视频会议室。[27]
A resource can be declared a Stoa by tagging it as such. [25] The main consequence of declaring a resource as a Stoa as of the time of writing is that the Agora will consider the resource as embeddable by default, and will try to provision it in relevant contexts. The Agora of Flancia also tries to provide useful default Stoas in every node: at least one public (currently world-writable) text document which allows for concurrent live editing, [26] and a video conferencing room. [27]
图 10.5 参考 Agora 中提供的默认 stoas。基于 URL 约定的“浅层”集成偏好使提供各种协作空间的成本保持在较低水平。
Figure 10.5 Default stoas provided in the reference Agora. A preference for “shallow” integrations based on URL conventions keeps the cost of providing a variety of collaboration spaces low.
请注意，提供的默认 Stoa 提供程序的 URL 可以从基本位置和当前节点名称中以低成本推断出来，因为支持它们的服务都在文档的 URL 中包含用户提供的标识符。只要其他提供程序支持类似的用户选择的 URL，添加对它们的支持成本就非常低。这种方法使交叉链接和其他集成的成本保持在相对较低水平 [28]，并且可以看作是高级 Agora 模式的一个例子，该模式使用轻量级文本约定来探索最终或尽力而为的协作。
Note that URLs for the provided default Stoa providers can be cheaply inferred from a base location and the current node name, as the services backing them all include user-provided identifiers in the URLs for documents. Adding support for other providers is very low cost provided they support similar user-chosen URLs. This approach keeps the cost of cross-linking and other integrations relatively low [28] and can be seen as an example of the high-level Agora Pattern of using lightweight textual conventions to explore eventual or best-effort collaboration.
设计原则
Design principles
Flancia 的参考 Agora 实现了一种基于现成组件的简单分布式架构，其中系统的服务和数据交换方面保持独立，仅通过 Agora Root（映射到文件系统中的共享路径）相互交互。
The reference Agora of Flancia implements a simple distributed architecture based on off-the-shelf components where the serving and data exchange aspects of the system are kept independent, interacting with each other only through an Agora Root (mapping to a shared path in a filesystem).
Agora Root 是包含 Agora 定义的 git 存储库，这意味着一份基本合约设定了 Agora 的基调和高级目标，以及一份要定期集成的数据源列表。
Agora Root is the git repository containing the Agora definition, meaning a base contract which sets the tone and high-level goals of the Agora, and a list of data sources to be recurrently integrated.
Agora Bridge 包含受支持数据源的连接器和导入器。用户控制的 git 存储库是默认数据源类型，但计划提供其他类型。
Agora Bridge contains connectors and importers for supported data sources. User controlled git repositories are the default data source type, but others are planned.
Agora Server 包含一个支持查询、集成、节点渲染的 Web UI；并提供 JSON、RSS、RDF 端点。
Agora Server contains a web UI supporting querying, integration, node rendering; and providing JSON, RSS, RDF endpoints.
图 10.6 三向参考 Agora 设计：Agora Bridge 将来自存储库和流的资源提供给 Agora Root，之后它们可以被 Agora Server 集成并提供。
Figure 10.6 The three-way reference Agora design: Agora Bridge feeds resources from repositories and streams into an Agora Root, after which they can be integrated and served by Agora Server.
Flancia 的 Agora Root 是 Agora Commons 的一个示例种子，本章将全面介绍 Agora Commons。这意味着它试图提供代码和说明，以便从头开始启动一个可用的知识共享。这个根具体是一个 Git 存储库 [29]，截至撰写本文时，它包含以下内容：
The Agora of Flancia’s Agora Root is an example seed for an Agora Commons as the one described in this chapter in all its facets. This means it tries to provide code and instructions to bootstrap a working Knowledge Commons from scratch. This root is concretely a Git repository [29] which as of the time of writing contains:
sources.yaml 文件包含要集成到 Agora 的所有存储库的列表，以及其他有用信息，如附加集成端点和备用服务位置。
A sources.yaml file containing a list of all repositories to be integrated into the Agora, plus other useful information like additional integration endpoints and alternative serving locations.
CONTRACT.md 文件包含代表 Agora 高级目标和价值观的断言列表，兼作 Commons 中的行为准则和 Agora 协议的默认定义。
A CONTRACT.md file containing a list of assertions representing the high-level goals and values of the Agora, doubling as code of conduct and default definition of Agora Protocol in the Commons.
README.md 文件包含如何使用上述信息和免费软件配置 Agora 的说明。
A README.md file containing instructions on how to provision an Agora using the information above and free software.
相同的文件可以存在于任何用户提供的存储库中，在这种情况下，它们被视为自愿提供的，目的是（a）扩展和补充 Agora 的根定义 [30]；（b）宣布用户打算参与社区的意图和价值观。
The same files can be present in any user provided repository, in which case they are taken to be volunteered with the intention to (a) extend and complement an Agora’s root definitions [30]; (b) announce the user’s intents and values with which they intend to participate in the community.
Agora 协议
Agora Protocol
本节描述了一套惯例，用于公开商定一套价值观和意图，这些价值观和意图可以说是 Agora 的定义，并在此后进行清晰的沟通。最简单的定义是，Agora 是任何将自己定义为公共空间并遵循一套明确的惯例以促进沟通和合作的公共空间。我们将这套惯例称为 Agora 的 Agora 协议，这里的协议既指“一套用于传达信息的规则和指南”（与本段中的含义最一致），也指“一套在特定情况下采取什么行动的程序”（因为 Agora 平台可以解析通过 Agora 协议表达的意图并采取相应的行动）。
This section describes a set of conventions for publicly agreeing on the set of values and intents that can be said to define an Agora and communicate clearly in it thereafter. In the barest of definitions an Agora is any public space that defines itself as such and follows an explicit set of conventions meant to facilitate communication and cooperation. We refer to this set of conventions as the Agora Protocol of an Agora, with protocol here meaning both “a set of rules and guidelines for communicating information” (aligning most clearly with the meaning in this paragraph) and “a set of procedures for what actions to take in a certain situation” (as the Agora Platform may parse intents expressed through Agora Protocol and take actions accordingly).
治理
Governance
Flancia 的 Agora 采取的方法是尝试在其根合约（如上一节所述）中使用自然语言定义一致的 Agora 协议，并从那里开始传递。 此后，它假定其用户仍然同意此根合约，对彼此的合约感兴趣，并愿意容忍观点和价值观的部分差异，特别是那些不会直接妨碍朝着共同目标合作或只是暂时的差异。 始终鼓励公开讨论合约及其互动的细节。 从高层次上讲，Flancia 的 Agora 根合约可以概括为呼吁社区保持诚信，遵循慈善原则并牢记言论的三道门。 [31] 平民或 Agorans 应根据需要和兴趣定期审查公共合约的变更，并在默认情况下相互互动时保持合理和友善。
The Agora of Flancia takes the approach of trying to define a consistent Agora Protocol using natural language in its root contract (described in the previous section) and transitively from there. It thereafter assumes that its users remain in agreement with this root contract, are interested in each other’s contracts, and are willing to tolerate partial discrepancies in views and values, in particular those not directly impeding cooperation towards shared goals or only temporary in nature. Open discussion about details of contracts and their interactions is encouraged at all times. At a high level, the Agora of Flancia’s root contract could be summarized as an appeal to the community to assume good faith, follow the Principle of Charity and be mindful of the Three Gates of Speech. [31] Commoners or Agorans are expected to periodically review changes to public contracts according to need and interest, and be reasonable and kind when interacting with each other by default.
Flancia 的 Agora 目前几乎没有制定任何应对对抗性的规定，并且在存储库被纳入其源列表后不提供访问控制系统：默认情况下，所有节点实际上都是集体所有。这意味着 Agorans 可能会向某些节点发送垃圾邮件，要么直接贡献大量资源，要么向它们推送（有关推送的更多信息见下文）；或者以其他方式提交可能被社区其他成员视为共享空间违规的内容。冲突解决机制预计将由社区进一步设计和开发，遵循运作良好的 Commons 治理的既定原则，并编码在 Agora 协议中。
The Agora of Flancia currently makes few provisions to account for adversariality of any kind, and provides no access control system after a repository has been admitted to its list of sources: all nodes are in effect collectively owned by default. This means that Agorans might spam certain nodes, either by directly contributing many resources or pushing to them (more on push below); or otherwise commit what might be perceived by the rest of the community as infractions in shared space. Conflict resolution mechanisms are expected to be designed and developed further by the community, following established principles of governance of well-functioning Commons, and encoded in Agora Protocol.
如果对来源或 Agora 协议的定义存在分歧，Agora 可能会通过分叉变成两个或多个。如果通过了分叉 Commons 的提案，那么将产生两个具有自洽定义的 Agora；如果可能的话，上述 Agora 有望在分叉后继续在同一个 Agora 网络内相互联合。对称地，Agora 有望偶尔合并成一个更大的 Commons，因为它们会探索共享空间并找到价值观和目标的一致性。治理（投票）和联合（互操作）机制的细节有望在 Agora 中得到足够详细的定义，但除此之外，这些内容超出了本章的范围。
In the case of disagreements about the definition of sources or Agora Protocol, an Agora might become two or more by forking. The result of a proposal to Fork the Commons, if it passes, would be two Agoras with self-consistent definitions; said Agoras are expected to continue to federate with each other post-fork within the same Agora Network, if possible. Symmetrically, Agoras are expected to occasionally Merge into a greater Commons as they explore shared space and find values and goals alignment. The particulars of Governance (voting) and Federation (interop) mechanisms are expected to be defined in an Agora in sufficient detail but are otherwise considered beyond the scope of this chapter.
文本惯例
Textual conventions
当以文本格式进行通信时，Agora 协议最具体地是一系列可选的印刷约定，允许用户贡献、链接或注释资源，而不管书写媒介是什么。如果您正在阅读本书，那么您可能在阅读之前就至少了解某种形式的 Agora 协议；它旨在反映个人知识管理领域的现有实践，并以默认协作的立场选择性地扩展它们。如果您使用 Tool for Thought 或以其他方式记下相互关联的笔记，并且可以在公共位置将这些笔记导出为 Agora 支持的格式，那么您已经谈论了协议的核心，并且从技术上讲，您已经拥有 Agora 编辑器（更多信息见下文）。
When communicating in text format, the Agora Protocol is most concretely a series of optional typographical conventions that allow users to contribute, link or annotate resources regardless of the writing medium. If you are reading this book, you probably know at least some form of Agora Protocol even before reading about it; it is meant to reflect existing practices in the Personal Knowledge Management space, and extend them optionally with a default-collaborative stance. If you use a Tool for Thought or otherwise take interlinked notes, and you can export these to a format supported by the Agora in a public location, you already talk the core of the protocol, and you technically already have an Agora Editor (more on that below).
例如，在贡献给 Agora 的文本资源中，[[Wikilinks]]（双方括号内的文本）可用于在您贡献的任何文章中以长格式指定实体和类别。当需要对特定的无损印刷实现进行编码时，Wikilinks 是一个不错的选择，即使在没有明确支持它们的编辑器和平台中，它们仍然清晰可读，这使它们成为轻量级跨工具互连约定的合理默认选择。也可以选择使用主题标签，根据需要使用 PascalCase 或 camelCase [32] 来编码句子；它们具有原生支持和社交媒体平台中预先存在的使用的优势。
In text resources contributed to an Agora, for example, [[Wikilinks]] (text within double square brackets) can be used to designate entities and categories in long form in any of your contributed writing. Wikilinks are a good fit when meaning to encode a particular lossless typographical realization, and they remain legible even in editors and platforms that don’t implement explicit support for them, which makes them a reasonable default choice for a lightweight cross-tool interlinking convention. Hashtags can optionally also be used, using PascalCase or camelCase [32] as needed to encode sentences; they have the advantage of native support and preexisting usage in social media platforms.
请注意，大纲加上标记和链接似乎足以以人类可读的方式对任意复杂度的思想和结构进行编码，同时还便于解析。如前所述，Logseq 等工具和参考文献 Agora 中探讨的默认基于树的方法似乎可以有效地以人性化和计算机友好的方式对超图进行编码和传达。这种方法可能会部分消除为实现 [[块引用]] 或包含而以编程方式生成 ID 的需要。[33] 这反过来又降低了此类机制的联合成本，因为系统不需要就共享 ID 方案达成一致。有关示例应用程序，请参阅下面有关操作 Push 的部分。
Note that outlines plus tagging and linking seem sufficient to encode thoughts and structure of arbitrary complexity in a human readable way, while also facilitating parsing. As mentioned earlier, a default tree-based approach as explored in tools like Logseq and in the reference Agora seems efficient to encode and communicate a hypergraph in a human friendly and computer friendly way. Such an approach might do away partially with the need to programmatically generate IDs for the purpose of implementing [[block references]] or transclusion. [33] This in turn lowers the cost of federation for such mechanisms, as there is no need for systems to agree on a shared ID scheme. See the section on action Push below for an example application.
从更一般的层面上讲，如果某人在公共场合宣布共享资源位于 Agora 中，则此后该资源可被视为位于该 Agora 中，除非 Agora 中的社区反对、声明被撤回或资源被发现与现有合同不兼容。Flancia 的 Agora 可以乐观地认为，贡献者打算使用 Agora 协议的变体有效地传达结构化含义；因此，该人打算在 Commons 中以默认的合作意图表达其个人目标和价值观。可以说，只要参与方明确同意这种意图并留下公开（或 Commons 可访问）的记录，Agora 和 Agora 协议的这种纯粹传统的定义可能会扩展到面对面的交谈、一般的写作和广泛的思想。[34]
On a more general level, if a person declares in a public setting that a shared resource is in an Agora, thereafter the resource may be considered to be in said Agora by definition – unless the community in the Agora objects, the statement is retracted, or the resource is found to be incompatible with existing contracts. The Agora of Flancia may optimistically assume that the person contributing means to communicate structured meaning efficiently using a variation of Agora Protocol; and accordingly, that the person means to express their individual goals and values in the Commons with a default intent of cooperation. This purely conventional definition of the Agora and Agora Protocol could be said to potentially extend to face-to-face conversations, to writing in general, and to thought at large, as long as participating parties explicitly agree on such an intent and leave a public (or Commons-accessible) record. [34]
Agora 操作
Agora Actions
Agora Actions 既是 Agora 平台功能，也是 Agora 协议的应用。Agora Action 可以通过写入任何贡献给 Commons 的资源来触发。方法是用 #tagging 或 [[wikilinking]] 标记块或资源，并根据上下文传递参数。检测到此意图后，Agora 平台将尝试执行该操作。Flancia 的 Agora 目前实施以下操作，作为其向社区提供普遍有用的功能的一部分。
Agora Actions are both Agora Platform features and an application of Agora Protocol. An Agora Action can be triggered by writing in any resource contributed to the Commons. This is done by #tagging or [[wikilinking]] a block or resource with the name of the action and passing it parameters by context. On detecting this intent, an Agora Platform will try to execute the action. The Agora of Flancia currently implements the following actions as part of its intent to provision generally useful functionality to its community.
去
Go
#go 或 [[go]] 是一个 Agora 动作，它将 URL 指定为在块出现的节点上下文中的规范或排名良好的 URL，并在执行时重定向到规范的 URL。
#go or [[go]] is an Agora Action that designates URLs as canonical or well ranked in the context of the nodes in which the block appears, and redirects to a canonical URL when executed.
Go 链接提供了一个有趣的案例研究，说明社区如何在知识共享环境中提供有用的服务。Go 链接最初是在硅谷公司开发和使用的，但很可能有潜力传播开来，并作为互联网规模的社交服务提供更广泛的效用。简而言之，Go 链接被标记为社交书签 - 由实践社区中的用户与 URL 关联的字符串。Go 链接作为自助服务命名重定向出现：提供它们所需的只是一个键/值存储和一个可以从所选标签（例如，[[go/agora]]）重定向到目标 URL（例如，https://flancia.org/agora）的 Web 服务。
Go links provide an interesting case study of how a community can provision useful services in the context of a Knowledge Commons. Go links were first developed and used in Silicon Valley corporations but likely have potential to spread and provide wider utility as an internet-scale social service. Put simply, Go links are labelled social bookmarks – strings associated with URLs by users in a community of practice. Go links emerged as self-service named redirects: all that is needed to provision them is a key/value store and a web service that can redirect from a chosen label (e.g., [[go/agora]]) to a target URL (e.g., https://flancia.org/agora).
作为简单（广泛）和复合（据我所知，特定于 Agora）Go 链接的示例，请考虑节点 [[pkg book]] 中的以下块：
As examples of both simple (widespread) and composite (to the best of my knowledge, Agora specific) Go links, consider the following blocks in node [[pkg book]]:
#前往 https://personalknowledgegraphs.com/
#go https://personalknowledgegraphs.com/
#推特 https://twitter.com/PkgBook
#twitter https://twitter.com/PkgBook
这些将导致 Flancia 的 Agora 将查询 [[go/pkg book]]（执行 Go 操作）重定向到 https://personalknowledgegraphs.com/ 并将 [[go/pkg book/twitter]] 重定向到 https://twitter.com/PkgBook。如果用户尝试解析 Go 链接并发现它缺失，Flancia 的 Agora 将提供关联的节点上下文（例如，上例中的 [[pkg book]]）并尝试设置一个流程，让用户可以在其他地方执行搜索，解析链接并在需要时将其返回到系统。
These will result in the Agora of Flancia redirecting query [[go/pkg book]] (an execution of the Go action) to https://personalknowledgegraphs.com/ and [[go/pkg book/twitter]] to https://twitter.com/PkgBook. If a user tries to resolve a Go link and finds it missing, the Agora of Flancia will serve the associated node context (e.g., [[pkg book]] in the previous example) and try to set up a flow where the user can perform a search elsewhere, resolve the link and contribute it back to the system if needed.
一个驱动假设是 Go 链接是有用的社会认知产物。[35] 在 Go 链接丰富的环境中，用户可以依赖其他用户在他们之前为命名实体定义规范链接；也就是说，在 Commons 中共享的与当前术语特别相关的资源集合通常可以“猜出”，即乐观地检索。这减少了查找和访问给定实体的规范资源的阻力。
One driving hypothesis is that Go links are useful social cognitive artifacts. [35] In a Go link rich environment, users can depend on other users to have defined canonical links for named entities before them; that is, collections of especially relevant resources to the terms at hand, as shared in a Commons, can often be “guessed”, i.e., retrieved optimistically. This reduces the friction to finding and visiting canonical resources for a given entity.
更一般地说，Go 链接代表了许多好的 Agora Actions 的样子：它们提供（a）有用的单独功能，因为它们允许用户将感兴趣的资源与（通常是短的）字符串关联起来，然后从任何具有网络访问权限的计算机快速调用该资源；并且它们还提供（b）以低成本或无需额外成本的社交功能，因为默认情况下共享链接（在命名空间中）可显着降低在需要时共享相关资源的摩擦。
More generally, Go links are representative of what many good Agora Actions might look like: they provide (a) a useful individual function, as they allow the user to associate a resource of interest with a (usually short) string, and then quickly recall the resource from any computer with network access; and they also provide (b) a social function at low or no additional cost, as sharing links (in a namespace) by default significantly lowers the friction of sharing relevant resources when needed.
拉
Pull
#pull 或 [[pull]] 是 Flancia 的 Agora 实现的另一个 Agora Action。其效果是一种 [[transclusion]] 形式：pull 将导致传递的资源或节点嵌入当前上下文中。它还向 Agora 暗示传递的实体与当前节点有某种对用户有意义的密切关系，这可能会对排名算法和等价类定义产生影响。
#pull or [[pull]] is another Agora Action implemented by the Agora of Flancia. Its effect is a form of [[transclusion]]: pulling will result in the passed resources or nodes being embedded in the current context. It also hints to an Agora that the entity passed is strongly related to the current node in some way meaningful to the user, which might have effects in ranking algorithms and equivalence class definitions.
#pull 接受一个 [[node]]、一个由两个节点组成的组合 [36]（形式为 [[node0/node1]]）或者一个 URL。
#pull takes a [[node]], a composition [36] of two nodes in the form [[node0/node1]], or a URL.
#拉 [[foo]]
#pull [[foo]]
将在下面嵌入（包含）节点 [[foo]]。
Will embed (transclude) node [[foo]] below.
#拉 [[@alice/bar]]
#pull [[@alice/bar]]
将仅嵌入用户@alice 贡献的[[bar]]中的子节点。
Will embed only the subnodes in [[bar]] contributed by user @alice.
#拉 [[foo/bar]]
#pull [[foo/bar]]
将在 [[foo]] 中嵌入标有 [[bar]] 的块或部分。
Will embed blocks or sections tagged with [[bar]] in [[foo]].
#拉取 https://example.org
#pull https://example.org
如有可能，将嵌入目标 URL。[37]
Will embed the target URL if possible. [37]
图 10.7 拉动操作。#pull 包含当前上下文中的目标节点或资源。
Figure 10.7 Pull in action. #pull transcludes the target node or resource in the current context.
推
Push
#push 或 [[push]] 是另一个 Agora Action。其效果也可以描述为包含，但方向与 #pull 相反：Pull 将在当前上下文中包含远程上下文，而 Push 将在远程上下文中包含当前上下文。Push 可以被认为类似于在 PubSub 系统或主题映射系统中发布到主题，主题映射到传递的 Agora 节点。
#push or [[push]] is another Agora Action. Its effect can also be described as transclusion, but in the opposite direction to #pull: whereas Pull will transclude a remote context in the current context, Push will transclude the current context in a remote context. Push can be thought of as similar to publishing to a topic in a PubSub system or topic map system, with the topic mapping to the Agora node that is passed.
#push 以节点、节点组合或要推送到的 URL [38] 作为参数；以及一些要推送的块。简而言之，推送到节点相当于在目标节点中写入推送的块或上下文；参考 Agora 以与本地资源类似的格式将它们发布到目标位置，根据配置和节点形状，要么在它们之前，要么在它们之后。此外，目标节点可能指示推送的插入点，在这种情况下，推送可能会完全包含在特定的远程资源中。
#push takes as parameter a node, a node composition, or a URL [38] to push to; and some blocks to be pushed. In short, pushing to a node is equivalent to writing the pushed blocks or context in the destination node; the reference Agora publishes them in the target location in a similar format to local resources, either preceding them or after them depending on configuration and node shape. Additionally, the target node might indicate an insertion point for pushes, in which case the push may be fully transcluded in a particular remote resource.
图 10.8 推送操作。操作 #push 远距离嵌入，这意味着它将当前资源或块嵌入到远程上下文中（如果可能）。在这里，您可以看到目标节点中来自不同来源的两次推送的结果。
Figure 10.8 Push in action. Action #push transcludes at a distance, meaning it embeds the current resource or blocks in a remote context (if possible). Here you can see the result of two pushes from different sources in the destination node.
推送到节点组合（看起来像“节点路径”）是请求将资源或块附加到远程位置的特定插入点，如果找到匹配的锚点，则意味着找到一个锚点（基本上允许对位置进行微调）。
Pushing to a node composition (which looks like a “node path”) is a request to attach the resource or blocks at a particular point of insertion in the remote location, meaning an anchor if a matching one is found (essentially allowing for fine tuning of placement).
请注意，在 Agora 中，即使没有明确提及推送操作，部分标题中的链接也可能将下一节推送到所提及的节点。此类行为称为“自动推送”，其发生的程度取决于 Agora 和用户偏好。
Note that in an Agora, links in section headings may push the following section to the mentioned nodes even without an explicit mention of the push action. Such behavior is termed “auto push”, and the degree to which it takes place is based on Agora and user preferences.
弗朗西亚集市的目标之旅
A tour of the goals of the Agora of Flancia
本节探讨了参考 Agora 正在进行的实验以及更大的 Agora Commons 的可能应用。本部分以一系列简短的探索性文章的形式进行，涵盖了当前和未来的工作。
This section explores experiments being undertaken with the reference Agora and possible applications of a greater Agora Commons. This is done in the form of a series of short exploratory essays covering present and future work.
截至 2023 年初，[39] Flancia 的 Agora 拥有一个新生但绝对有趣的社区和相关的 Agora Graph：60 位用户（加上另外约 100 位来自社交媒体的贡献者）贡献了 35k 个资源，Agora 将这些资源整合为 25k 个节点，在它们之间建立了 127k 个链接。
As of early 2023, [39] the Agora of Flancia hosts a nascent but definitely interesting community and associated Agora Graph: 60 users (plus another ~100 contributing from social media) have contributed 35k resources that the Agora integrates as 25k nodes, setting up 127k links between them.
这意味着参考 Agora 是一个小型 Agora，因此它只能开始测试这里概述的一些假设。Flancia Collective 负责提供和维护此 Agora，旨在促进和支持创建更大更好的 Agora，广义上是指有利于亲社会协调与合作的公共和半公共环境。这里的更好意味着更易于使用、更易于访问，并且通常对越来越多样化的 Agorans 更有用。具体来说，这意味着 Flancia 的 Agora 的目标之一是继续开发参考 Agora 平台，以便技术背景较少的人可以使用（并享受）Agora Commons。
This means the reference Agora is a small Agora, and as such it can only begin to test some of the hypotheses sketched here. Provisioning and maintenance of this Agora is undertaken by Flancia Collective with the intention to promote and support the creation of greater and better Agoras, defined widely as public and semi-public environments conducive to pro-social coordination and cooperation. Here better means easier to use, more accessible, and generally more useful to an ever-greater diverse set of Agorans. Concretely this means that the Agora of Flancia has amongst its goals continuing to develop the reference Agora Platform so that people with less technical backgrounds can make use of (and enjoy) an Agora Commons.
如果 Agora 继续作为一项实验发展，请注意当前的参考实现可能无法很好地垂直扩展，并且需要进行架构升级才能轻松支持更大的实例。这一事实与上一段中表达的意图相结合，导致 Flancia Collective 专注于通过联合网络方法实现水平扩展。这种方法还具有以下好处：(a) 与 Fediverse 中已经广泛使用的去中心化方法兼容；(b) 有机地导致更多种类的 Agora，其中一些可以是特定于主题或学科的，这可能会引起人们对新社区的兴趣，并允许尝试特定于社区的功能，从而提高每个实例的用户友好度。
If the Agora continues to grow as an experiment, note that the current reference implementation will likely not scale well vertically, and architectural upgrades will be needed to comfortably support much larger instances. This fact, combined with the intention expressed in the previous paragraph, lead to Flancia Collective focusing on achieving scaling horizontally through a federated network approach. This approach also has the benefit of (a) being compatible with the decentralized approach already used widely in the Fediverse and (b) leading organically to a greater variety of Agoras, some of which can be topic- or discipline-specific, which may raise interest in new communities and allow for experimenting with community-specific features increasing user friendliness on a per instance basis.
上述网络方法正在 https://agor.ai 中进行试验，其根目录为参考 Agora 提供服务，并将按照类似 subreddit 的方案在其子域中为各种特定主题的 Agora 提供服务。例如，在此方案中，https://pkg.agor.ai 可能是一个专门包含与个人知识图谱相关的节点的 Agora，也可能是一个此类节点排名上升或优先显示的 Agora。
The above network approach is being trialed in https://agor.ai, which serves the reference Agora in its root, and will serve a variety of topic-specific Agoras in its subdomains following a subreddit-like scheme. In this scheme, for example, https://pkg.agor.ai may be an Agora exclusively containing nodes related to Personal Knowledge Graphs, or one where such nodes are upranked or surfaced preferentially.
利用开放算法探索新兴意义
Exploring emergent meaning with open algorithms
Flancia 的 Agora 试图根据所谓的 Agorans 或 Commoners 提供的信号，为社交和知识空间中的常见问题提供开放式解决方案。其中一些信号可能是从 Agora 平台的使用中推断出来的，例如，当用户浏览节点并贡献注释时，而其他信号可能是明确传达的，并以用户使用 Agora 协议宣布的意图的形式出现。
The Agora of Flancia tries to provision open solutions to common problems in the social and knowledge spaces based on signals contributed by so-called Agorans or Commoners. Some of these signals might be inferred from use of the Agora Platform, for example, as users browse nodes and contribute annotations, while others might be explicitly communicated and take the shape of intents announced by users using Agora Protocol.
更具体地说，Flancia 的 Agora 试图提供跨工具、跨平台的服务，使社区能够探索排名、过滤和协调问题的解决方案。驱动假设是，最终的意义、目标和价值观的融合将来自分布式和异构的构建块，作为 Commons 中的一种新兴现象，并使 Agorans 和更广泛的社区受益。
More concretely, the Agora of Flancia tries to provision cross-tool, cross-platform services enabling a community to explore solutions to ranking, filtering, and coordination problems. The driving hypothesis is that eventual convergence on meaning, goals, and values will arise from distributed and heterarchical building blocks as an emergent phenomenon in the Commons, and benefit both Agorans and the wider community.
考虑一个 Agora，它可能会通过逐步收敛的过程为节点定义等价类。首先，让 D(x, y) 成为 Agora Graph 中节点 x 和 y 之间的距离，按照一些有用的度量，默认情况下是 Agora 图中从 x 到 y 的路径长度（跳跃为链接）。D(x, y) < 1 可能是考虑到节点之间的进一步亲和性提示而得出的，例如 Agora 协议操作（如一个节点拉动另一个节点）或自然语言提示（如两个节点在附近的区块中交替提及）。考虑一个基本系统，其中每个额外的提示都会将先前已知的节点之间的距离减半。然后，如果在聚合（社交）层面上 x 和 y 有足够的亲和性提示将它们联系起来，Agora 可能会形成一个基本信念，即 x ~ y，这意味着 x 和 y 在上下文中是一致的。假设是，随着越来越多的用户贡献亲和性信号，这将成为现实，从反映等价性的有用定义的意义上来说。这可以形象地理解为两个节点在图中任意靠近，在超过用户选择的半径后就变成等价的。
Consider an Agora which might define equivalence classes for nodes through a process of incremental convergence. First, let D(x, y) be the distance between nodes x and y in the Agora Graph as per some useful metric, by default the length of the path from x to y in the Agora graph (with hops being links). D(x, y) < 1 can result from taking into account further affinity hints between nodes, for example Agora Protocol actions like one node pulling the other, or natural language hints like the two nodes being mentioned alternatively in nearby blocks. Consider a basic system in which each additional hint halves the previously known distance between nodes. Then, if on an aggregated (social) level x and y have enough affinity hints linking them, an Agora might develop a base belief that x ~ y, meaning x and y are congruent within a context. The hypothesis is that this will become true, in the sense of reflecting a useful definition of equivalence, as more and more users contribute affinity signals. This might be visualized as the two nodes becoming arbitrarily close in the graph, snapping to equivalence after crossing a user-chosen radius.
举个例子：在任何一个集市中，按原样提及的节点 [[Marx]] 可能意味着 [[Karl Marx]] 或 [[Groucho Marx]]。在 Flancia 集市中，Karl Marx 与节点 Marx 之间存在“相互拉动”关系，因此只要这样做有成效，就会默认将其视为等价关系（例如，在选择如何对节点内的子节点进行排序、如何解决传递拉动，或在本地上下文图中显示哪些链接时）。
As an example: in any given Agora, node [[Marx]] mentioned as is could mean alternatively [[Karl Marx]] or [[Groucho Marx]]. In the Agora of Flancia, Karl Marx is in a “mutual pull” relationship with node Marx, so it will be treated as equivalent by default whenever it is productive to do so (for example when choosing how to rank subnodes within a node, how to resolve transitive pulls, or which links to display in the local context graph).
扩展这种方法，Flancia 的 Agora 打算在 Commons 中实现纯开放的排名和过滤算法。链接在社交环境中的共享方式为相关性、显著性甚至潜在的病毒式传播提供了强烈的信号。[40] 根据我的经验，在 Agora 中，对于任何给定的 Agoran，其他 Commoners 共享的链接和资源往往默认很有趣。这两个观察结果让我相信，一些 Agora 应该发现开发由社交信号支持的内部高质量排名和过滤程序是可行的。此外，请注意，Go 链接已经促进了 URL 与描述性标签的共享，这为训练推荐系统或通用搜索引擎提供了良好的语料库。动作 #pull 也可以用作排名提示：如果 x 拉动 y，y 中的资源可能与 x 的上下文相关，因此默认情况下可能会出现在节点 [[x]] 中。这也适用于在 [[y]] 等中拉动的节点，具有衰减因子。每当 Agora 拉取“过多”时，用户可以通过添加和删除拉取或限定拉取资源来优化拉取的资源集。
Extending this approach, the Agora of Flancia intends to implement purely open ranking and filtering algorithms in the Commons. How links are shared in social contexts provides a strong signal for relevance, notability, and even potential virality. [40] It is also my experience that within an Agora, to any given Agoran, links and resources shared by other Commoners tend to be interesting by default. These two observations lead me to believe that some Agoras should find it tractable to develop in-house high-quality ranking and filtering procedures supported by social signals. In addition, note that Go links already promote the sharing of URLs alongside descriptive labels, which makes for a good corpus for training a recommendation system or general-purpose search engine. Action #pull can also be used as a ranking hint: if x pulls y, resources in y are likely to be relevant in contexts about x, and may thus be surfaced in node [[x]] by default. This applies transitively also to nodes pulled in [[y]], etc. with an attenuation factor. Whenever an Agora pulls “too much,” users might refine the set of resources pulled by adding and removing pulls or qualifying them.
要探索的基本方法是聚合人类可读的信号，以引导一组透明的模型和开源算法，帮助社区融合思维模型、管理 Commons 并规范与其他系统的交互。任何给定的 Agora 都应支持替代算法；理想情况下，排名和过滤算法应该是即插即用的，最终可以在客户端实现。
The basic recipe to be explored is that of aggregating human-readable signals to bootstrap a set of transparent models and open source algorithms that assist the community in converging on mental models, governing the Commons and regulating interactions with other systems. Alternate algorithms should be supported in any given Agora; ideally, ranking and filtering algorithms should be plug and play and could eventually be implemented client-side.
集中我们的共同注意力
Focusing our shared attention
在活跃期间，Agora 社区每周或每月都会商定一个节点，作为松散并发写作活动的最佳连接点。我们将这一过程称为“节点俱乐部”，它有助于我们作为一个社区找到共同的关注焦点。
Once a week, or month, while active, the Agora community agrees on a node to serve as a best-effort nexus of loosely concurrent writing activity. We call this process, which facilitates finding a shared focus for our attention as a community, Node Club.
弗朗西亚集市节点俱乐部的创始人将其描述为“每周在集市中进行的公共漂流”（Mather，2021 年）。他还在弗朗西亚集市的著作中进一步探讨了集市作为公共空间的潜力：例如，参见节点弗朗西亚集体和公共空间领域。
The founder of the Node Club of the Agora of Flancia describes it as “a week-by-week communal drift through the Agora” (Mather, 2021). He has also further explored the potential of the Agora as a Commons in his writing in the Agora of Flancia: see for example node Flancia Collective and the Spheres of Commoning.
此次集体点头活动产生的一些节点列在 Flancia 集市的匹配节点中：https://anagora.org/node-club。
Some nodes that resulted from this communal nodding activity are listed in the matching node in the Agora of Flancia: https://anagora.org/node-club.
维护数字自决权
Upholding the right to digital self-determination
Agora 试图在用户所在的地方与他们会面。根据这一愿景，Flancia 的 Agora 专注于为社区提供工具，以便他们根据需要将自托管数据导入和导出知识共享，以及建立和维护与一般有趣的数字空间（包括围墙花园）之间的桥梁。
An Agora tries to meet the user where they are. In line with this vision the Agora of Flancia focuses on providing the community with tools for importing and exporting their self-hosted data in and out of the Knowledge Commons as they wish, and for building and maintaining bridges to and from interesting digital spaces in general, including walled gardens.
在 Flancia 的 Agora 中，要成为正式用户，目前需要联系管理员[41]，告知他们所需的用户名、要导入的存储库以及对基本 Agora 合同的接受。[42] 请注意，简单的分布式设计依赖于人们使用他们独立的 Agora 编辑器向优先自托管的存储库写入内容，这确保用户可以继续使用他们选择的本地工具来查看和编辑他们自己的个人知识图谱，这保证了可访问性按照每个 Agora 编辑器实施的标准进行，并且个人数据在 Flancia 的 Agora 或任何 Agora 继续存在之后仍然可用。
In the Agora of Flancia, joining as a full user currently involves reaching out to a steward, [41] informing them of a desired username, a repository to be imported and the acceptance of the base Agora contract. [42] Note the simple distributed design, which relies on people using their independent Agora Editors to write to preferentially self-hosted repositories, ensures that users can continue to use local tools of their choosing to view and edit their own Personal Knowledge Graphs, which guarantees accessibility as per the standard implemented by each individual Agora Editor, and personal data availability beyond the continued existence of the Agora of Flancia, or indeed any Agora.
话虽如此，这个过程并不是很全面，因为它需要一些设置和（截至撰写本文时）相当程度的技术知识。为了让人们更轻松地将个人资源链接到 Agora 并选择性地写入，Flancia 的 Agora 运行了一组社交媒体账户，感兴趣的各方可以关注这些账户，以贡献帖子，这些帖子将从各种来源链接到 Commons。[43] 用户还可以选择加入完整消息写入，这样每条消息的完整副本都会包含一个 wikilink 被存储。如果用户尚未设置 Agora 存储库，则帖子内容将以每个用户一个存储库的方案保存在服务器端，目的是让用户尽快控制他们的存储库。
Having said that, this process is not very inclusive, as it requires some setup and (as of the time of writing) a fair degree of technical know-how. To let people link and optionally write individual resources to the Agora more easily, the Agora of Flancia runs a set of social media accounts that interested parties can follow to contribute posts to be linked into the Commons from a variety of sources. [43] Users can additionally opt in to full message writing, which results in a full copy of each message containing a wikilink being stored. If the user has not yet set up an Agora repository, post content is kept server-side with a one-repository-per-user scheme, with the intent of letting users take control of their repositories as soon as possible.
引导通用问题解决装置
Bootstrapping a general problem-solving device
Flancia 的 Agora 乐观地寻求提供和维护一个 Agora Graph，该图谱旨在为各种问题建模并最终解决这些问题。目前，该知识图谱包含其自身及其相关社区的描述，以及在相关问题解决环境中有用的模式存储库的托管或链接。
The Agora of Flancia optimistically seeks to provision and maintain an Agora Graph tailored to the goal of modeling, and eventually solving, a variety of problems. Currently this knowledge graph contains a description of itself and its associated community, and hosts or links to repositories of patterns useful in relevant problem solving contexts.
作为一个合作团体，Agorans 默认采用一种天真但建设性的方法解决问题。具体如下。对于社区感兴趣的每个问题：尝试在 Agora 内尽可能全面地描述它。作为这项工作的一部分，维护一组已知的候选解决方案。对于每个解决方案：尝试将其描述为一个过程（一种算法）和一个依赖关系图，对实现它所需的资源（时间、注意力、财富）进行建模。
Agorans, as a cooperative group, are asked to adopt a naive but constructive approach to problem solving by default. This is as follows. For each problem interesting to the community: try to describe it as thoroughly as possible within the Agora. As part of this effort, maintain a set of known candidate solutions. For each solution: try to describe it as a procedure (an algorithm) and a dependency graph modeling the resources needed to implement it (time, attention, wealth).
在进行上述操作时，参与用户可能会声明他们的世界源 (M) 和目标 (M′) 模型，例如使用 Agora 协议以及社区同意交换的任何通用格式。这使他们能够清楚地定义他们认为最有趣或值得优先考虑的问题和解决方案集。同意集合和优先级的用户可以通过汇集公共资源来有效地协作解决方案。驱动假设是，在具有默认合作立场的群体中促进清晰的沟通和明确的行为可能会导致群体功能和相关问题解决能力的逐步改善，这些改善会随着时间的推移而累积。
While undertaking the above, participating users might declare their source (M) and target (M′) models of the world, for example using Agora Protocol alongside any common formats that the community agrees to exchange. This results in them clearly defining which problem and solution sets they consider maximally interesting or worthy of prioritization. Users that agree on sets and priorities can then efficiently collaborate on solutions as they become available by the pooling of common resources. The driving hypothesis is that the promotion of clear communication and explicit behavior in groups with a default-cooperative stance may result in unlocking incremental improvements in group functioning and associated problem solving, which compound over time.
假设人类未来存在一种有效的通用问题解决设备或类似的认知神器，弗兰西亚的 Agora 开始尝试引导这种设备的最小可行版本并促进有关它的讨论。事实证明，将构建更好的 Agora 作为 Agora 的基本问题是一种不可抗拒的诱惑。也就是说，参考 Agora 试图探索构建一个系统所涉及的问题，该系统允许人们和团体相互理解并最佳地合作解决问题，即使面对偏见和非理性等障碍也是如此。[44] 您正在阅读的文本是这个过程的产物之一。
Assuming the existence of an efficient generalized problem-solving device or similar cognitive artifact in the future of humanity, the Agora of Flancia was started to try to bootstrap a minimum viable version of such a device and promote discussion about it. It then proved irresistible to seed the Agora with the base problem of building a better Agora. That is, the reference Agora tries to explore precisely the problems involved in building a system that allows people and groups to understand each other and cooperate optimally to solve problems even in the face of obstacles like biases and irrationality. [44] The text you are reading is one of the products of this process.
结束语
Closing remarks
在本章中，我试图展示一个社区如何仅使用广泛可用的思考工具、共享的真相来源和明确的默认合作立场来探索各种项目。在这样做的过程中，我一次又一次地依赖参考 Agora 提供的示例，它是 (Graph、Platform、Protocol) 三元组的一个特定实例，可以说是定义了 Agora。参考实现的历史和现在可能不如 Agora 项目背后的总体愿景和原则那么重要，但在我们分道扬镳之前，我想分享更多导致开发这个特定 Agora 的背景和动机。
In this chapter I have tried to show how a community might set out to explore a variety of projects using little more than widely available tools for thought, a shared source of truth, and an explicit default cooperative stance. While doing so I have relied again and again on the examples offered by the reference Agora, which is one particular instantiation of the (Graph, Platform, Protocol) tuple that can be said to define an Agora. The history and the present of the reference implementation might not be as important as the general vision and principles behind the Agora project, but before we part ways, I would like to share a bit more of the background and motivations that led to developing this particular Agora.
需要在其母项目的背景下评估 Flancia 集市的一些特征。Flancia (https://flancia.org) 是我在 2018 年空闲时间启动的一个项目，旨在探索利他主义和原托邦主义，最初是在虚构的框架内。Flancia 最初将集市仅作为一种叙事手段，解释了一个社会如何能够可靠地解决各种各样的协调问题并超越这些问题，实现更大的利益。几年后，Flancia Collective 是一个松散但绝对真实的群体，与 Flancia 项目有关，该项目为更广泛的社区提供了非虚构的参考集市 (https://anagora.org)。
Some of the characteristics of the Agora of Flancia need to be evaluated in the context of its parent project. Flancia (https://flancia.org) is a project I started in my free time back in 2018 to explore Altruism and Protopianism, initially within a fictional framing. Flancia originally included the Agora only as a narrative device which explained how a society had been able to reliably solve a wide variety of coordination problems and advance beyond them towards a greater good. Fast forward a few years and Flancia Collective is a loose but decidedly real group of people associated with the Flancia project which provisions the non-fictional reference Agora (https://anagora.org) for the benefit of a wider community.
在整个过程中，弗兰西亚不断发展，并将其部分影响逐渐带入阿戈拉。在政治和哲学层面，弗兰西亚可能最显著地受到模式语言（作者未提及）《Bolo'Bolo》（Widmer，1983 年）的影响，这是无政府主义者对超越资本主义和共产主义后的世界的重新思考；以及《扩展的圈子》（Singer，2011 年），它连接了社会生物学和伦理学领域，为有效利他主义奠定了基础，弗兰西亚就是在这种背景下诞生的。阿戈拉作为非暴力革命思想和教育的平台，直接得益于默里·布克钦在《21 世纪革命》（Bookchin，2015 年）中发展了市政主义，以及保罗·弗莱雷的里程碑式著作《被压迫者教育学》（Freire，2014 年）。所有这些工作结合在一起，最紧密地体现了分布式利他主义、更广泛的民主参与、财富再分配和社区驱动的问题解决的理想，而弗兰西亚集市希望在与之互动的人们和团体中推广和实现这些理想。
All through this process Flancia has continued to develop and bring some of its influences transitively into the Agora. On the political and philosophical dimensions Flancia is informed maybe most notably by the pattern language (not named as such by its author) Bolo’Bolo (Widmer, 1983), an anarchist rethinking of the world as it could be after evolving beyond Capitalism and Communism; and by The Expanding Circle (Singer, 2011), which bridges the fields of sociobiology and ethics and laid much of the groundwork for Effective Altruism, in the context of which Flancia was born. The Agora as a platform for nonviolent revolutionary thought and education is directly indebted to Murray Bookchin for developing Municipalism in The Revolution in the 21st Century (Bookchin, 2015) and Paulo Freire for his landmark work Pedagogy of the Oppressed (Freire, 2014). All these works taken together most closely inform the ideals of distributed altruism, greater democratic participation, wealth redistribution and community-driven problem solving that the Agora of Flancia aspires to promote and enable in the people and groups it interacts with.
话虽如此，请随意忽略或至少质疑本章中提到的但与您自己的不一致的 Flancia 集会背后的目标或信念。集会是关于探索多元化的力量，我希望你的声音有一天可以加入一些类似的善意合唱团，即使不是完全参考集会。集会项目的真正潜力在于其未来的多样性。
Having said that, please feel free to disregard or at least question the goals or beliefs behind the Agora of Flancia which were mentioned throughout this chapter but aren’t aligned with your own. The Agora is about exploring strength in plurality, and I hope your voice can someday join, if not exactly the reference Agora, some similar well-meaning Chorus of Voices. The true potential of the Agora project is to be found in its future diversity if anywhere at all.
Flancia 和 Flancia 的 Agora 是知识共享和开源的。请考虑分叉、重新混合和/或将反馈转发给 Agora 的管理员。就此而言，我现在将留给你们可以称为 Flancia 的 Agora 的官方副歌：
Flancia and the Agora of Flancia are Creative Commons and Open Source. Please consider forking, remixing and/or forwarding feedback to the stewards of the Agora. In this vein I will now leave you with what could be called the official refrain of the Agora of Flancia:
如果你不喜欢这个 Agora，请建造一个更好的！
If you don’t like this Agora, please build a better one!
谢谢
Thanks
我要感谢 Flancia Collective、更广泛的 Agora 社区以及 Fellowship of the Link：感谢你们的启发、贡献和指导。
I would like to thank Flancia Collective, the wider Agora Community, and the Fellowship of the Link: for your inspiration, contributions and guidance.
我要感谢我的朋友们：感谢你们的爱、支持和理解。
I would like to thank my friends: for your love, support and understanding.
我要感谢历史上集市的建设者和维护者，包括那些建造历史城邦和我们城市的公共空间的人，以及那些将建设未来更好的集市的人。
I would like to thank the Agora builders and maintainers through history, including those who built the public spaces of the historical poleis and our cities, and those who will build the better Agoras of the future.
我想感谢您阅读本文。
I would like to thank you: for reading this.
祝你幸福！
May you be happy!
笔记
Notes
[1] https://anagora.org/@flancian (Agora)，https://social.coop/@flancian (Fediverse)。
[1] https://anagora.org/@flancian (Agora), https://social.coop/@flancian (Fediverse).
[2] 古斯塔夫森（2020 年）。
[2] Gustafson (2020).
[3] 如果节点 A 中的资源包含 wikilink [[B]] 或主题标签 #B，则 A 链接到 B。根据 Agora 默认设置和用户偏好，直接提及实体名称也可能导致节点之间建立链接。有关更多信息，请参阅节点 https://anagora.org/wikilinks-everywhere。
[3] If a resource in node A contains a wikilink [[B]] or hashtag #B, A is linked to B. Plainly mentioning an entity by name might also result in a link between nodes depending on Agora default and user preferences. For more on this, please refer to node https://anagora.org/wikilinks-everywhere.
[4] 可在 https://anagora.org 和 https://flancia.agor.ai 上获取。
[4] Available on https://anagora.org and https://flancia.agor.ai.
[5] 可从 https://anagora.org/go/agora-server、https://anagora.org/go/agora-bridge 获取。有关技术概述，请参阅本章附录，可从 https://anagora.org/go/agora-chapter-appendix 获取。
[5] Available at https://anagora.org/go/agora-server, https://anagora.org/go/agora-bridge. For a technical overview, see the appendix to this chapter available at https://anagora.org/go/agora-chapter-appendix.
[6] 及其所有子域，子域由同一 Agora 网络中的不同实例提供服务。
[6] And all its subdomains, subdomains being served by different instances in the same Agora network.
[7] 包含各种社交图谱，既有显性的（例如，有关人及其关系的条目），也有隐性的（例如，在社会背景下产生的知识；个人或社区承载着社会信息的文化和艺术表现）。
[7] Contains a variety of social graphs, both explicit (e.g., entries about people and their relationships) and implicit (e.g., knowledge produced in a social context; cultural and artistic manifestations by individuals or a community which carry social information).
[8] 从技术上讲，你可以将其视为相互链接的节点元组，或者更一般地，将其视为通过复合关系（例如带有注释的链接）相互关联的节点集或节点序列。后者会产生超图 - 而这正是 Agora 通常试图实现和提供的。
[8] Technically you can think of it as a tuple of nodes linking to each other, or more generally a set or sequence of nodes related to each other via a composite relationship (e.g., a link with annotations). The latter results in a hypergraph – which is what an Agora generally tries to implement and provision.
[9] 弗兰西亚集市的知识共享空间是为了公众利益和所有众生的利益而设立的。
[9] The Knowledge Commons of the Agora of Flancia is provisioned for the public good and the benefit of all beings.
[10] 奥斯特罗姆，E.（1990 年）。
[10] Ostrom, E. (1990).
[11] Hess 和 Ostrom（2007 年）。
[11] Hess and Ostrom (2007).
[12] 亚历山大，C.（1977 年，1979 年）。
[12] Alexander, C. (1977, 1979).
[13] Bollier 和 Helfrich（2019 年）。
[13] Bollier and Helfrich (2019).
[14] 或者不这样做。目前，Agora 中的突发行为大多是一个未经检验的假设。有关这方面的更多信息，请参阅解决难题部分。
[14] Or do not. At this point in time, emergent behavior in the Agora is mostly an untested hypothesis. For more on this, please refer to the section Tackling hard problems.
[15] 所提供的 Agora Bridge 目前可以导入联合 wiki，并最终允许 Agora 与 FedWiki 完全互操作。
[15] The provided Agora Bridge can currently import federated wikis and should eventually allow an Agora to fully interop with FedWiki.
[16] 如需查看本章图表的更新版本及其他版本，请参阅https://anagora.org/agora-resources。
[16] To see updated versions of diagrams in this chapter and additional ones, please refer to https://anagora.org/agora-resources.
[17] 初始存储库设置后，Agora Bridge 会轮询更新。新资源或更新的资源通常会在提交后几秒钟内由 Agora Server 提供，模数缓存设置。
[17] After initial repository setup, Agora Bridge polls for updates. New or updated resources are usually served by Agora Server within a few seconds of commit, modulo caching settings.
[18] 有关超图的更多信息，请参阅 Fong 和 Spivak (2019) 和 https://ncatlab.org/nlab/show/hypergraph。
[18] For more on hypergraphs see Fong and Spivak (2019) and https://ncatlab.org/nlab/show/hypergraph.
[19] Joslyn 和 Nowak（2018 年）。
[19] Joslyn and Nowak (2018).
[20] 它本身就是一个可以访问和写入的节点；Agora 不会对链接和节点进行严格区分。特别是，Flancia 的 Agora 将链接建模为节点 3 元组（源、注释、目标），这恰好非常适合 RDF。有关 RDF 的良好入门知识，请参阅 https://anagora.org/go/rdf-primer。
[20] Which is in itself a node that can be visited and written about; the Agora doesn’t make heavy distinctions between links and nodes. In particular, the Agora of Flancia models links as a node 3-tuple (source, annotation, destination), which incidentally fits RDF quite nicely. For a good primer on RDF, please see https://anagora.org/go/rdf-primer.
[21] 默认情况下，[[Wikilinks]] 和 #hashtags 被视为等效，选择其中一个取决于方便程度或风格偏好。我个人倾向于对单个单词使用 #tags，对多词短语使用 wikilinks。
[21] [[Wikilinks]] and #hashtags are considered equivalent by default, and choosing one or the other is a matter of convenience or stylistic preference. I personally tend to use hashtags for single words and wikilinks for multi-word phrases.
[22] Agora 尝试在 Commons 中提供符合道德的搜索服务。
[22] An Agora tries to provision ethical search services in the Commons.
[23] 请参阅 https://anagora.org/users 中的各个条目。
[23] See individual entries in https://anagora.org/users.
[24] 事实上，根据我的经验，在 Agora 中找到我想要写的笔记是使用 Agora 最令人高兴的事情之一。
[24] Indeed, finding notes that I meant to write already written in the Agora is one of the most joyous aspects of using it, in my experience.
[25] 在这里和其他地方，标记意味着在链接前加上标签或维基链接，并导致 Agora 图中出现标记边。
[25] Here and elsewhere, tagging means prefixing a link with a hashtag or wikilink, and results in a labeled edge in the Agora graph.
[26] 截至撰写本文时，Hedgehoc 或 Etherpad。
[26] As of the time of writing, Hedgehoc or Etherpad.
[27] 截至撰写本文时，Jitsi Meet。
[27] As of the time of writing, Jitsi Meet.
[28] 截至撰写本文时，Agora 和 Stoa 提供商之间还没有通信，但可以做出进一步的规定（例如）同步权限和在 API 级别交换数据。
[28] As of the time of writing there is no communication between the Agora and Stoa providers, but further provisions could be made to (for example) synchronize permissions and exchange data at the API level.
[29] https://github.com/flancian/agora。
[29] https://github.com/flancian/agora.
[30] 递归解析源可能会产生递归 Agoras:) Agora 递归的另一种方式是由于每个单独的节点都可以看作是由节点中的子节点及其交互定义的子图，因此本质上 Agora 无论如何都是 Agoras 的组合。
[30] Resolving sources recursively may yield recursive Agoras :) The other way in which the Agora is recursive is due to the fact that each individual node can be seen as a subgraph defined by the subnodes in the node and their interactions, so in essence an Agora is in any case a composition of Agoras.
[31] Linvega（2022 年）。
[31] Linvega (2022).
[32] 帕斯卡大小写是指标签（或编码中的变量）中去掉空格，每个单词的首字母大写：ThisUsesPascalCase。骆驼大小写与之类似，但只有第一个单词的首字母是小写的：thisUsesCamelCase。
[32] Pascal case is where a hashtag (or variable in coding) has the spaces removed and an uppercase first letter of each word: ThisUsesPascalCase. Camel case is similar, but the first letter of just the first word is lowercase: thisUsesCamelCase.
[33] 我认为，就 [[认知人工制品]] 而言，[[块引用]] 具有相对竞争力，因为实现往往需要数据库驱动的创建和检索，以及更普遍的计算不透明的解析程序。 Agora 操作加上人类可读的节点 ID 可以说是互补的，因为参与系统（包括人类）可以使用描述性标签来建模关系，这是一种强化机制。
[33] I believe that [[Block References]] are relatively competitive as far as [[Cognitive Artifacts]] go, as implementations tend to require database-driven creation and retrieval, and more generally computationally opaque resolution procedures. Agora actions plus human-readable node IDs can be said to be complementary as participating systems (including humans) can model relationships using descriptive labels, which works as a reinforcement mechanism.
[34] 如此广泛的 Agora 定义是否有用且是否具有持久的模因力还有待观察。
[34] It remains to be seen whether such a wide-ranging definition of an Agora is useful and has staying memetic power.
[35] 依据 Krakauer（2016）和 Norman（1991）的观点。
[35] As per Krakauer (2016) and Norman (1991).
[36] 在参考文献中，Agora 斜线组合 (slash-composition) 被实现 (默认情况下) 作为投影操作，这意味着 [[foo/bar]] 会将 [[foo]] 中的资源或块过滤为提及 [[bar]] 的资源或块。其他组合运算符正在计划中，但截至撰写本文时尚未实现。
[36] In the reference Agora slash-composition is implemented (by default) as a projection operation, meaning that [[foo/bar]] will filter resources or blocks in [[foo]] to those mentioning [[bar]]. Other composition operators are planned but not implemented as of the time of writing.
[37] 这受 X-Frame-Options 和 Content-Security-Policy 规定的网站和浏览器安全政策约束。
[37] This is subject to site and browser security policies as per X-Frame-Options and Content-Security-Policy.
[38] 尚未实施；预计依靠 [[Webmentions]] 来告知目标有关推送和其他 Agora 交互的信息。
[38] Not implemented yet; expected to rely on [[Webmentions]] for informing targets about pushes and other Agora interactions.
[39] 最新统计数据可参见https://anagora.org/nodes。
[39] Up-to-date statistics can be found at https://anagora.org/nodes.
[40] 截至本文撰写时，社交网络已提供充足的证据。例如，请参阅 Bu 等人 (2010)。
[40] Social networks provide ample evidence of this as of the time of writing. See for example Bu et al (2010).
[41] 有权修改 Agora 真相来源的人。
[41] Someone with access to modify an Agora’s sources of truth.
[42] 另外，有技术头脑的潜在普通人可以向真相来源发送拉取请求，将自己添加到根存储库。有关最新程序，请参阅 https://anagora.org/join。
[42] Alternatively, the tech-minded prospective commoner can send a pull request to the source of truth adding themselves to the root repository. For up-to-date procedures, please refer to https://anagora.org/join.
[43] 截至撰写本文时，请在 Fediverse 上关注 @agora@botsin.space 或在 Twitter 上关注 @an_agora。有关社交和聊天网络上可用的系统帐户的最新列表，请参阅 https://anagora.org/agora-bot。
[43] As of the time of writing, follow @agora@botsin.space in the Fediverse or @an_agora on Twitter. For an up-to-date list of system accounts available on social and chat networks, please refer to https://anagora.org/agora-bot.
[44] 有关该问题领域的更多信息，请参阅知识未来小组 (https://www.knowledgefutures.org/)、开放全球思维 (https://openglobalmind.com/) 和规范辩论实验室 (https://canonicaldebatelab.com/) 等的工作。有关该领域相关项目的最新列表，请参阅 https://anagora.org/sense-making。
[44] For more in this problem space please refer to the work of the Knowledge Futures group (https://www.knowledgefutures.org/), Open Global Mind (https://openglobalmind.com/) and Canonical Debate Lab (https://canonicaldebatelab.com/) among others. For an up to date list of relevant projects in this space please refer to https://anagora.org/sense-making.
后记
Afterword
亲爱的读者，
Dear reader,
您已到达个人知识图谱之旅的终点。感谢您与我们同行。当我们出发时，我们对这本书寄予厚望，并设定了很高的标准。
You’ve reached the end of this personal knowledge graph journey. Thank you for sailing along with us. When we set out, we had high hopes and set high standards for this book.
我们希望这本书通俗易懂、内容丰富。因此，我们汇集了多方投稿，为各行各业的投稿人和读者提供了广阔的空间。我们还决定将这本书本身打造成个人知识图谱，使用领域内的工具和实践来编写和发布它。
We wanted this book to be accessible and inclusive. This is why we compiled the book from multiple contributions and opened the scope for contributors and readers from all walks of life. We also decided to make this book a personal knowledge graph itself, using tools and practices from the domain to write and publish it.
这样做并不容易，但它让我们沉浸在一种我们原本不会拥有的独特体验中。现在这段旅程已经结束，我们的期望得到了满足吗？值得吗？
Doing so was not easy, but it immersed us in a unique experience we would not otherwise have. Now that this journey has reached its end, were our expectations met? Was it worth it?
我们探索了个人知识图谱的过去、现在和未来。知识图谱有着学术背景，行业用例也正在蓬勃发展。连接这两个世界本身就够难的了。对于个人知识图谱，还需要让工具极其直观，并面临拥有不同背景和需求的用户的额外挑战。
We explored the past, present and potential future of personal knowledge graphs. Knowledge graphs have an academic background and use cases in the industry are burgeoning. Bridging these worlds is hard enough as it is. For personal knowledge graphs, there is the added need to make tools extremely intuitive, and the added challenge of having users with diverse backgrounds and needs.
“个人知识图谱”对不同的人来说有不同的含义。然而，对大多数人来说，这个术语的定义是松散而模糊的。对大多数人来说，它归结为一种他们可以使用的工具，主要用于做笔记，但也用于做许多其他事情。
“Personal knowledge graphs” means different things to different people. For most people, however, the term is loosely and vaguely defined. For most people, too, it comes down to having a tool they can use, mainly to take notes, but also to do a number of other things.
个人知识图谱工具生态系统既生机勃勃又难以驾驭。工具供应商来来去去，瞬息万变，优先级也瞬息万变。工具功能千差万别。工具互操作性严重不足。这些是我们设计一个框架来评估不同的工具并使用其中许多工具共同创作这本书时得出的主要结论。
The personal knowledge graph tool ecosystem is lively as much as it is unruly. Tool vendors come and go and change priorities in a flash. Tool capabilities vary wildly. And tool interoperability is sorely lacking. Those were our key findings from devising a framework to evaluate different tools and using a number of those to collaboratively create this book.
鉴于这些事实，我们可以理解为什么没有一家更广泛使用的工具供应商响应我们的征集。尽管我们很想与世界分享创造者塑造这个生态系统未来的想法，但我们必须假设他们更重视构建功能和用户群。
Given those facts, we can understand why none of the more widely used tool vendors answered our call for submissions. As much as we would have liked to share the thoughts of makers shaping the future of this ecosystem with the world, we’ll have to assume they give building features and userbases higher priority.
然而，这并不是 PKG 工具领域的全部。探索多模式和人际图表让我们充满希望，ImageSnippets 和 Agora 为各自社区带来的功能可以传播并被更多工具采用。
That, however, is not all there is to the PKG tool landscape. Exploring multimodal and inter-personal graphs left us hopeful that the facilities that ImageSnippets and Agora bring to their respective communities could spread and be adopted by more tools too.
为了使用户能够无缝地从一种工具迁移到另一种工具，我们必须再次强调互操作性的必要性。虽然每种 PKG 工具在存储方面都有自己的选择，但拥有一种用于数据交换的中间格式将实现互操作性。Markdown 可以成为这种格式，但它需要标准化和丰富化。
For that to happen in a way that enables users to migrate seamlessly from one tool to the other, we must reemphasize the need for interoperability. While each PKG tool makes its own choices in terms of storage, having an intermediate format for data exchange would enable interoperability. Markdown could be that format, but it needs to be standardized and enriched.
我们也尝试使用 Markdown 来编写这本书，但效果并不理想。我们的想法是围绕 Markdown 存储库（在本例中是在 GitHub 上）构建协作，该存储库可用于跟踪贡献和编辑。虽然 Markdown 运行良好，但协作依赖于第三方工具集成。这效果不佳，不得不被搁置一旁。也许 SamePage 最终会实现这一目标。
We tried using Markdown for authoring this book as well, but that did not work for us. The idea was that collaboration would be structured around a Markdown repository, in this case on GitHub, that could be used to keep track of contributions and edits. While Markdown and work well, collaboration relied on third-party tool integrations. That did not work well and had to be sidelined. Perhaps SamePage will get there eventually.
一个正交但相关的问题是能够表达个人知识图谱的语义。这将提升连接在个人知识图谱工具中作为一等公民的地位，因为可以表达它们的含义和方向。为此使用知识图谱世界中的 RDF 标准将是一个合理的选择，我们探讨了一些如何做到这一点的建议。
An orthogonal but related issue is that of being able to express semantics for personal knowledge graphs. That would promote the status of connections as first-class citizens in personal knowledge graph tools, as it would be possible to express their meaning and direction. Using the RDF standard from the world of knowledge graphs for this would be a reasonable choice, and we explored some proposals for how this could be done.
我们还探讨了一项提案，该提案可以在 SEN、桌面 PKG 应用程序和一些可跨工具利用的可用性模式上使用不同的词汇。
We also explored a proposal that would enable using different vocabularies on SEN, a desktop PKG application, and a few usability patterns that can be leveraged across tools.
个人知识图谱的前景如何？
What lies ahead for Personal Knowledge Graphs?
问题有待解决，但潜力巨大。一如既往，正是我们自身知识和 PKG 社区之间的联系，让我们的船队能够绕过暗礁，在低迷时期继续前进，并驶向地平线，驶向生产力、创造力和意外发现的新海洋。
There are issues to solve but massive potential to release. As always, it’s the connections, both within our own knowledge and between us as a PKG community, that will see our flotilla navigate around the rocks, keep us moving in the doldrums and see us cruise over the horizon to new oceans of productivity, creativity and serendipity.
我们想再次感谢您与我们一起航行。
We want to thank you once more for sailing along with us.
如果你想加入我们，可以在这里找到我们：https://PersonalKnowledgeGraphs.com
And if you want to come aboard then you can find us here: https://PersonalKnowledgeGraphs.com
编辑们
The editors
伊沃·维利奇科夫、乔治·阿纳迪奥蒂斯
Ivo Velitchkov, George Anadiotis
Exapt Press 的更多内容
More from Exapt Press
请访问 https://exapt.press 来查找更多关于以不同方式看待世界的书籍。
Go to https://exapt.press to find more books on seeing the world differently.
Exapt Press 专门出版复杂性和系统思维方面的书籍。这些书籍将世界视为一个整体，并为旧问题和情况带来不同的观点或新的分析。
Exapt Press specialises in complexity and systems thinking books. Books that see the world as a whole and bring a different viewpoint or new analysis to old problems and situation.
订阅我们的时事通讯，即可定期收到新书公告、新闻和书评。
Sign up to our newsletter to receive regular book announcements, news, and book reviews.
在 Twitter 上关注我们：@ExaptPress 和@Rob_Worth（主编）。
Follow us on Twitter: @ExaptPress and @Rob_Worth (Editor-in-Chief).
Ivo Velitchkov 的《基本平衡》
Essential Balances by Ivo Velitchkov
您真的了解您的组织吗？
Do you really understand your organization?
除非您理解了基本平衡，否则您不会知道。
You don’t until you understand the Essential Balances.
阅读 Ivo Velitchkov 撰写的《基本平衡：停止观察，开始了解组织运作的原理》
Read Essential Balances: Stop Looking and Start Seeing What Makes Organizations Work by Ivo Velitchkov
更多信息请访问 https://exapt.press/essential-balances
Find out more at https://exapt.press/essential-balances
罗马俱乐部的极限与超越
Limits and Beyond from the Club of Rome
1972年，一本书改变了世界。
In 1972, a book changed the world.
《增长的极限》中的模型正确吗？
Were the models in The Limits to Growth right?
我们听了吗？
Did we listen?
过去 50 年发生了什么？
What happened in the last 50 years?
是不是太晚了？
Is it too late?
更多信息请访问 https://exapt.press/limits-and-beyond
Find out more at https://exapt.press/limits-and-beyond
关于贡献者
About the Contributors
George Anadiotis – 联合编辑
George Anadiotis – Co-editor
由于拥有 IT 背景，他有机会学习演奏多种乐器，并最终成为一名单人乐队和管弦乐指挥。
Coming from an IT background, he’s had the chance to learn to play many instruments on the way to becoming a one man band and an orchestrator.
他曾担任过软件工程师、架构师、顾问、经理、研究员和分析师。他建立和管理过各种规模和类型的项目、产品和团队，做过屡获殊荣的研究，创办过初创公司，为财富 500 强企业、初创公司和非政府组织提供服务，并且活了下来。
He has worked as a Software Engineer, Architect, Consultant, Manager, Researcher and Analyst. He has built and managed projects, products and teams of all sizes and shapes, done award-winning research, founded startups, served Fortune 500 companies, startups and NGOs, and lived to tell the tale.
他是 Linked Data Orchestration 和 Year of the Graph 的创始人，也是 Connected Data World 的董事总经理。George 还担任 GigaOm 分析师，并且是 VentureBeat 和 ZDNet 的撰稿人。
He is the Founder of Linked Data Orchestration and the Year of the Graph, and the Managing Director of Connected Data World. George also works as a GigaOm Analyst and is a VentureBeat and ZDNet Contributor.
乔治认为 PKG 生态系统的出现是实现个人层面数字主权和人际层面网络协作的推动因素。知识图谱为人民服务！
George sees the emergence of a PKG ecosystem as an enabler towards digital sovereignty on a personal level, and networked collaboration on an interpersonal level. Knowledge graphs for the people!
Ivo Velitchkov – 联合编辑
Ivo Velitchkov – Co-editor
Ivo 是一位独立的管理和数据顾问。在过去的 27 年里，他一直与大型公共和私人组织合作，帮助他们制定战略、结构和信息。他曾担任项目经理、首席执行官、教练、顾问、研究员和培训师。他是《Essential Balances》一书的作者，也是《企业架构和互联电子政务：实践与创新》一书的合著者。Ivo 还撰写博客：https://www.strategicstructures.com。他拥有计算机科学博士学位。
Ivo is an independent management and data consultant. For the last 27 years, he has worked with big public and private organizations helping them with their strategy, structures and information. He has worked as a project manager, CEO, coach, consultant, researcher, and trainer. He’s the author of the book Essential Balances, and co-author of the book Enterprise Architecture and Connected E-Government: Practices and Innovations. Ivo also writes a blog: https://www.strategicstructures.com. He has a PhD in Computer Science.
Ivo 已经培训了数百人如何使用和应用语义技术，并帮助大型组织将他们的数据统一到企业知识图中。除了开放和企业知识图谱之外，在过去几年中，他还积极致力于推广个人知识图谱的采用。
Ivo has trained hundreds of people how to use and apply semantic technologies and has helped large organizations unify their data in enterprise knowledge graphs. Apart from open and enterprise knowledge graphs, in the last couple of years, he’s been actively working to spread the adoption of personal knowledge graphs.
玛丽贝尔·阿科斯塔
Maribel Acosta
Maribel Acosta 是波鸿罗尔大学的助理教授，她是该校数据库和信息系统小组的负责人，也是神经计算研究所 (INI) 的成员。她拥有卡尔斯鲁厄理工学院 AIFB 研究所知识管理小组的博士学位。在她的科学生涯中，她开发了结合语义网、数据库和知识表示的形式化技术来解决知识图谱管理的不同问题。她的研究重点是将知识图谱的符号表示与知识图谱构建、查询和完成的方法相结合。
Maribel Acosta is an Assistant Professor at the Rhur-University Bochum, where she is the Head of the Database and Information Systems group and a member of the Institute for Neural Computation (INI). She holds a PhD from the Knowledge Management group at the AIFB Institute, at the Karlsruhe Institute of Technology. In her scientific career, she has developed techniques that combine formalisms from the Semantic Web, Databases, and Knowledge Representation to address different problems of Knowledge Graph management. Her research focuses on combining symbolic representations of KGs with methods for KG construction, querying, and completion.
奥梅斯·巴尔特斯
Omes Baltes
Omes Baltes 是来自德国的软件工程师。他在波鸿罗尔大学学习数学和计算机科学，目前在 ObjectCode GmbH 担任 Web 开发人员。
Omes Baltes is a software engineer from Germany. He studies Math and Computer Science at the Rhur-University Bochum and works as a web developer at ObjectCode GmbH.
法布里斯·加莱
Fabrice Gallet
哲学老师。用作 PKG 并用于教育目的，开发一些 JS 扩展和 Smartblocks，同时学习（或者说重新学习）编码。总体上对思维工具以及任何可以帮助开发和分享更精确、更有启发性的想法的东西更感兴趣，尤其是论证图。
Philosophy teacher. Using as a PKG and for educational purposes, and developing some JS extensions and Smartblocks, while learning (or rather relearning) to code. More generally interested in Tools for Thought and anything that can help develop and share more precise and enlightening thoughts, especially argument maps.
爱德华多·伊万内茨 (弗朗西安)
Eduardo Ivanec (Flancian)
白天是具有 10 年经验的站点可靠性工程师、系统工程师；晚上是 Protopian 和知识图谱爱好者。
Site Reliability Engineer, Systems Engineer with 10 years of experience by day; Protopian and Knowledge Graph Aficionado by night.
马蒂纳斯·尤塞维丘斯
Martynas Jusevičius
Martynas 是知识图谱开发人员，负责从概念到实现。AtomGraph 联合创始人。坚信数据驱动架构、声明式技术和通用软件。他花了大量时间思考 Web 的第一原则。
Martynas is Knowledge Graph developer from concept to implementation. Co-founder of AtomGraph. Strong believer in data-driven architecture, declarative technologies and generic software. He has spent a good deal of time thinking about the first principles of the Web.
格雷戈尔·罗森瑙尔
Gregor Rosenauer
Gregor 拥有扎实的软件工程和敏捷方法基础，始终热衷于尝试新方法，使用最先进的工具、技术和最佳实践，基于开放标准提供创新且可靠的高质量解决方案。他的主要关注点是 Java 和服务/中间层的集成，尤其是 API 设计、技术/工具评估、原型设计、与利益相关者密切合作的产品开发。
Based on a solid foundation in software engineering and agile methods, Gregor is always eager to try out new ways to deliver innovative but reliable quality solutions based on open standards using state-of-the-art tools, technologies and best practices. His main focus is on Java and integration in the service/middle tier, especially API design, technology/tool evaluation, prototyping, product development in close alignment with stakeholders.
一些用来描述他的关键词：OSS 倡导者、伪装的创新者、父亲、自然爱好者；摄影、音频、电影和书籍爱好者（Tsundoku 诊断）。
Some keywords to describe him: OSS advocate, innovator in disguise, father, nature lover; photo, audio, cine- and bibliophile (Tsundoku diagnosis).
玛格丽特·沃伦
Margaret Warren
玛格丽特·沃伦是一名技术专家、研究员和艺术家。她是佛罗里达人类与机器认知研究所的研究员。她的研究涉及图像描述、形式化语义以及人类如何描述视觉体验。
Margaret Warren is a technologist, researcher and artist. She is a research associate with the Florida Institute of Human and Machine Cognition. Her research is related to image description, formalized semantics and how humans describe what they experience visually.
她为人类和机器创建了工具和流程，包括为机器学习优化数据集的管理、通过人工智能增强的图像描述任务的知识工程，以及结合人类和机器的专业知识将分类器和其他注释数据转换为多模态图。
She creates tools and processes for both humans and machines including the curation of optimized datasets for machine learning, knowledge engineering from image description tasks augmented by AI, and the transformation of classifiers and other annotation data into multi-modal graphs using a combination of human and machine expertise.
参考
References
Abiteboul, S. (1997)。查询半结构化数据。在 F. Afrati 和 P. Kolaitis (Eds.) 的《数据库理论 - ICDT '97》（第 1-18 页）中。Springer。https://doi.org/10.1007/3-540-62222-5_33
Abiteboul, S. (1997). Querying semi-structured data. In F. Afrati & P. Kolaitis (Eds.), Database Theory—ICDT ’97 (pp. 1–18). Springer. https://doi.org/10.1007/3-540-62222-5_33
Access Co., Ltd.（第二版）。Be Book—系统概述—应用程序套件。2022 年 9 月 28 日检索自 https://www.haiku-os.org/legacy-docs/bebook/TheApplicationKit_Scripting.html
Access Co., Ltd. (n.d.). The Be Book—System Overview—The Application Kit. Retrieved September 28, 2022, from https://www.haiku-os.org/legacy-docs/bebook/TheApplicationKit_Scripting.html
Adler, MJ 和 Doren, CV (1972)。《如何阅读一本书：智能阅读的经典指南》（修订版）。Touchstone。
Adler, M. J., & Doren, C. V. (1972). How to Read a Book: The Classic Guide to Intelligent Reading (Revised edition). Touchstone.
Ahrens, S. (2017)。如何记笔记：一种促进写作、学习和思考的简单技巧——适用于学生、学者和非小说类图书作家。CreateSpace 独立出版平台。
Ahrens, S. (2017). How to Take Smart Notes: One Simple Technique to Boost Writing, Learning and Thinking – for Students, Academics and Nonfiction Book Writers. CreateSpace Independent Publishing Platform.
Obenauer，A.（2021 年 8 月 21 日）。Graph OS。实验室笔记。https://alexanderobenauer.com/labnotes/014/
Obenauer, A. (2021, August 21). The Graph OS. The Lab Notes. https://alexanderobenauer.com/labnotes/014/
Alexander, C. (1979)。《形式综合笔记》（10 版）。哈佛大学出版社。
Alexander, C. (1979). Notes on the synthesis of form (10. print). Harvard University Press.
Alexander, C.、Ishikawa, S. 和 Silverstein, M. (1977)。模式语言：城镇、建筑、建设。牛津大学出版社。
Alexander, C., Ishikawa, S., & Silverstein, M. (1977). A pattern language: Towns, buildings, construction. Oxford University Press.
Alexopoulos, P. (2020)。数据语义建模：避免陷阱和打破困境（第一版）。O'Reilly Media。
Alexopoulos, P. (2020). Semantic modeling for data: Avoiding pitfalls and breaking dilemmas (First edition). O’Reilly Media.
Ames, A.、Maltzahn, C.、Bobb, N.、Miller, EL、Brandt, SA、Neeman, A.、Hiatt, A. 和 Tuteja, D. (2005)。使用链接和属性实现更丰富的文件系统元数据。第 22 届 IEEE/第 13 届 NASA Goddard 海量存储系统和技术会议 (MSST'05)，第 49–60 页。https://doi.org/10.1109/MSST.2005.28
Ames, A., Maltzahn, C., Bobb, N., Miller, E. L., Brandt, S. A., Neeman, A., Hiatt, A., & Tuteja, D. (2005). Richer File System Metadata Using Links and Attributes. 22nd IEEE / 13th NASA Goddard Conference on Mass Storage Systems and Technologies (MSST’05), 49–60. https://doi.org/10.1109/MSST.2005.28
Angles, R.、Arenas, M.、Barceló, P.、Hogan, A.、Reutter, J. 和 Vrgoč, D. (2018)。《图形数据库的现代查询语言基础》。ACM 计算调查，50(5)，1–40。https://doi.org/10.1145/3104031
Angles, R., Arenas, M., Barceló, P., Hogan, A., Reutter, J., & Vrgoč, D. (2018). Foundations of Modern Query Languages for Graph Databases. ACM Computing Surveys, 50(5), 1–40. https://doi.org/10.1145/3104031
Anytype——终身操作系统，（2023 年）。https://anytype.io/en#use-cases
Anytype—Operating system for life, (2023). https://anytype.io/en#use-cases
Ashby, WR (1991)。必要多样性及其对复杂系统控制的影响。载于 GJ Klir 的《系统科学的方方面面》（第 405-417 页）。Springer US。https://doi.org/10.1007/978-1-4899-0718-9_28（原作于 1958 年出版）
Ashby, W. R. (1991). Requisite Variety and Its Implications for the Control of Complex Systems. In G. J. Klir, Facets of Systems Science (pp. 405–417). Springer US. https://doi.org/10.1007/978-1-4899-0718-9_28 (Original work published 1958)
Ashby, WR (2014)。《为大脑而设计：适应性行为的起源》。Martino Fine Books。（原著出版于 1952 年）
Ashby, W. R. (2014). Design for a Brain: The Origin of Adaptive Behavior. Martino Fine Books. (Original work published 1952)
Ashby, WR (2015)。《控制论导论》。Martino Fine Books。（原著出版于 1956 年）
Ashby, W. R. (2015). An Introduction to Cybernetics. Martino Fine Books. (Original work published 1956)
雅典🏛 [@SDesearch]。（2022 年 11 月 11 日）。来自我们的 Discord 的更新：https://t.co/EdDvO20Dr2 [推文]。推特。https://twitter.com/AthensResearch/status/1591138491379122176
Athens 🏛 [@SDesearch]. (2022, November 11). Update from our Discord: Https://t.co/EdDvO20Dr2 [Tweet]. Twitter. https://twitter.com/AthensResearch/status/1591138491379122176
Athensresearch。（2022 年 11 月 11 日）。https://twitter.com/AthensResearch/status/1591138491379122176?s=20&t=8IZVzyr01LZhWj4e-KQSmQ
Athensresearch. (2022, November 11). https://twitter.com/AthensResearch/status/1591138491379122176?s=20&t=8IZVzyr01LZhWj4e-KQSmQ
Atomgraph HQ，V.（nd）。LinkedDataHub v3——关于 LinkedDataHub。LinkedDataHub。2022 年 9 月 29 日检索自 https://atomgraph.github.io/LinkedDataHub/linkeddatahub/docs/about/
Atomgraph HQ, V. (n.d.). LinkedDataHub v3—About LinkedDataHub. LinkedDataHub. Retrieved September 29, 2022, from https://atomgraph.github.io/LinkedDataHub/linkeddatahub/docs/about/
Axelrod, R. (2006)。合作的演变。Basic Books。
Axelrod, R. (2006). The evolution of cooperation. Basic Books.
Balog, K. 和 Kenter, T. (2019)。个人知识图谱：研究议程。2019 年 ACM SIGIR 信息检索理论国际会议论文集，217–220。https://doi.org/10.1145/3341981.3344241
Balog, K., & Kenter, T. (2019). Personal Knowledge Graphs: A Research Agenda. Proceedings of the 2019 ACM SIGIR International Conference on Theory of Information Retrieval, 217–220. https://doi.org/10.1145/3341981.3344241
Balog, K.、Mirza, P.、Wang, Z. 和 Skj, MG (2022)。AKBC 202 个人知识图谱研讨会 (PKG 2021) 报告。ACM SIGIR 论坛，56(1)，11。
Balog, K., Mirza, P., Wang, Z., & Skj, M. G. (2022). Report on the Workshop on Personal Knowledge Graphs (PKG 2021) at AKBC 202. ACM SIGIR Forum, 56(1), 11.
Baloo——KDE 社区 Wiki。 （2022 年 1 月 29 日）。 https://community.kde.org/Baloo
Baloo—KDE Community Wiki. (2022, January 29). https://community.kde.org/Baloo
Baty，J.（2022 年 10 月 10 日）。个人知识管理令人筋疲力尽——Baty.net。https://baty.net/2022/personal-knowledge-management
Baty, J. (2022, October 10). Personal Knowledge Management is exhausting – Baty.net. https://baty.net/2022/personal-knowledge-management
Benet, J. (2014)。IPFS – 内容寻址、版本化、P2P 文件系统 (arXiv:1407.3561)。arXiv。https://doi.org/10.48550/arXiv.1407.3561
Benet, J. (2014). IPFS – Content Addressed, Versioned, P2P File System (arXiv:1407.3561). arXiv. https://doi.org/10.48550/arXiv.1407.3561
Bergman, M.（2019 年 7 月 1 日）。知识图谱的常识性观点。AI3:::自适应信息。https://www.mkbergman.com/2244/a-common-sense-view-of-knowledge-graphs/
Bergman, M. (2019, July 1). A Common Sense View of Knowledge Graphs. AI3:::Adaptive Information. https://www.mkbergman.com/2244/a-common-sense-view-of-knowledge-graphs/
Bergman, O. 和 Whittaker, S. (2016)。《管理数字资料的科学》。麻省理工学院出版社。
Bergman, O., & Whittaker, S. (2016). The Science of Managing Our Digital Stuff. MIT Press.
Bergman, O.、Whittaker, S. 和 Schooler, J. (2021)。眼不见心不烦：书签被创建但未被使用。《图书馆学与情报科学杂志》，53(2)，338–348。https://doi.org/10.1177/0961000620949652
Bergman, O., Whittaker, S., & Schooler, J. (2021). Out of sight and out of mind: Bookmarks are created but not used. Journal of Librarianship and Information Science, 53(2), 338–348. https://doi.org/10.1177/0961000620949652
Berman, JJ (2022)。分类使科学家如何建立和使用分类和本体变得相关。学术出版社。
Berman, J. J. (2022). Classification made relevant how scientists build and use classifications and ontologies. Academic Press.
Bernardi, A.、Grimnes, G.、Groza, T. 和 Scerri, S. (2011)。 NEPOMUK 语义桌面（第 255-273 页）。 https://doi.org/10.1007/978-3-642-19510-5_13
Bernardi, A., Grimnes, G., Groza, T., & Scerri, S. (2011). The NEPOMUK Semantic Desktop (pp. 255–273). https://doi.org/10.1007/978-3-642-19510-5_13
Berners-Lee, T. (2009)。Tim Berners-Lee：下一个网络 | TED 演讲。https://www.ted.com/talks/tim_berners_lee_the_next_web
Berners-Lee, T. (2009). Tim Berners-Lee: The next web | TED Talk. https://www.ted.com/talks/tim_berners_lee_the_next_web
Berners-Lee, T. (2016 年 7 月 13 日)。巨型全球图 | 分散信息组 (DIG) 面包屑。https://web.archive.org/web/20160713021037/http://dig.csail.mit.edu/breadcrumbs/node/215
Berners-Lee, T. (2016, July 13). Giant Global Graph | Decentralized Information Group (DIG) Breadcrumbs. https://web.archive.org/web/20160713021037/http://dig.csail.mit.edu/breadcrumbs/node/215
Berners-Lee, T.、Hendler, J. 和 Lassila, O. (2001)。语义网。《科学美国人》，第 284 卷，第 5 期，第 34-43 页。
Berners-Lee, T., Hendler, J., & Lassila, O. (2001). The Semantic Web. Scientific American, Vol. 284, No. 5, 34-43.
Bhihe, C. (2022)。Cbhihe/findxattr [Shell]。https://github.com/Cbhihe/findxattr（原创作品于 2022 年发表）
Bhihe, C. (2022). Cbhihe/findxattr [Shell]. https://github.com/Cbhihe/findxattr (Original work published 2022)
块协议。（2022 年）。块协议：构建基于块的接口的开放标准。https://blockprotocol.org
Block Protocol. (2022). Block Protocol: The open standard for building block-based interfaces. https://blockprotocol.org
Bollier, D. 和 Helfrich, S. (2019)。自由、公平、生机：公地的叛乱力量。新社会出版社。
Bollier, D., & Helfrich, S. (2019). Free, fair, and alive: The insurgent power of the Commons. New Society Publishers.
Bookchin, M. (2015)。下一场革命：人民议会和直接民主的承诺。Verso Books。
Bookchin, M. (2015). The next revolution: Popular assemblies and the promise of direct democracy. Verso Books.
Bostrom, N. (2019)。脆弱世界假说。全球政策，10(4)，455-476。https://doi.org/10.1111/1758-5899.12718
Bostrom, N. (2019). The vulnerable world hypothesis. Global Policy, 10(4), 455-476. https://doi.org/10.1111/1758-5899.12718
Brander，G.（2021 年 7 月 23 日）。聚合器并非开放式的 [Substack 时事通讯]。潜意识。https://subconscious.substack.com/p/aggregators-arent-open-ended
Brander, G. (2021, July 23). Aggregators aren’t open-ended [Substack newsletter]. Subconscious. https://subconscious.substack.com/p/aggregators-arent-open-ended
Brander，G.（2022 年 8 月 19 日）。Noosphere，一种思想协议 [Substack 时事通讯]。潜意识。https://subconscious.substack.com/p/noosphere-a-protocol-for-thought
Brander, G. (2022, August 19). Noosphere, a protocol for thought [Substack newsletter]. Subconscious. https://subconscious.substack.com/p/noosphere-a-protocol-for-thought
Brickley, D. & Miller, L. (2004 年 8 月 18 日)。FOAF 词汇规范：命名空间文档 1 2004 年 5 月。http://xmlns.com/foaf/0.1/
Brickley, D. & Miller, L. (2004, August 18). FOAF Vocabulary Specification: Namespace Document 1 May 2004. http://xmlns.com/foaf/0.1/
Bu, J. 等人（2010 年）。“通过统一超图进行音乐推荐：
Bu, J. et al (2010). “Music Recommendation by Unified Hypergraph:
Bush, V. (1945)。正如我们所想。交互作用，3(2)，35–46。https://doi.org/10.1145/227181.227186
Bush, V. (1945). As we may think. Interactions, 3(2), 35–46. https://doi.org/10.1145/227181.227186
Bushnell, T. (2015 年 3 月 8 日)。《面向操作系统设计的新策略》，Thomas Bushnell, BSG 撰写的架构概述。https://www.gnu.org/software/hurd/hurd-paper.html#design
Bushnell, T. (2015, March 8). Towards a New Strategy of OS Design, an architectural overview by Thomas Bushnell, BSG. https://www.gnu.org/software/hurd/hurd-paper.html#design
Cadwalladr, C. 和 Graham-Harrison, E. (2018 年 3 月 17 日)。揭露：剑桥分析公司在重大数据泄露事件中收集了 5000 万个 Facebook 个人资料。卫报。https://www.theguardian.com/news/2018/mar/17/cambridge-analytica-facebook-influence-us-election
Cadwalladr, C., & Graham-Harrison, E. (2018, March 17). Revealed: 50 million Facebook profiles harvested for Cambridge Analytica in major data breach. The Guardian. https://www.theguardian.com/news/2018/mar/17/cambridge-analytica-facebook-influence-us-election
capacities.io。（2022 年 2 月 31 日）。Capacities.io。并非所有笔记都是平等的。https://capacities.io
capacities.io. (2022, February 31). Capacities.io. Not All Notes Are Created Equal. https://capacities.io
Cevolini, A. (2018)。Niklas Luhmann 的卡片索引从何而来？3，390–420。https://doi.org/10.1163/24055069-00304002
Cevolini, A. (2018). Where Does Niklas Luhmann’s Card Index Come From? 3, 390–420. https://doi.org/10.1163/24055069-00304002
Christl, W. (2017)。公司如何利用个人数据对付人们。自动化劣势、个性化说服以及个人信息商业使用的社会影响。https://scholar.google.com/citations?view_op=view_citation&hl=es&user=02RZEUYAAAAJ&citation_for_view=02RZEUYAAAAJ:IjCSPb-OGe4C
Christl, W. (2017). How Companies Use Personal Data Against People. Automated Disadvantage, Personalized Persuasion, and the Societal Ramifications of the Commercial Use of Personal Information.. https://scholar.google.com/citations?view_op=view_citation&hl=es&user=02RZEUYAAAAJ&citation_for_view=02RZEUYAAAAJ:IjCSPb-OGe4C
Clark, A. (2003)。天生的机器人：思想、技术和人类智能的未来。牛津大学出版社。
Clark, A. (2003). Natural-born cyborgs: Minds, technologies, and the future of human intelligence. Oxford University Press.
Clark, A., & Chalmers, D. (1998). 扩展思维。分析，58(1)，7-19。
Clark, A., & Chalmers, D. (1998). The Extended Mind. Analysis, 58(1), 7–19.
Colombetti, G. (2014).《感觉身体：情感科学与行动思维的交汇》。麻省理工学院出版社。
Colombetti, G. (2014). The Feeling Body: Affective Science Meets the Enactive Mind. MIT Press.
Colombetti, G. (2017)。《感觉身体：情感科学与行动思维的交汇》（重印版）。麻省理工学院出版社。
Colombetti, G. (2017). The Feeling Body: Affective Science Meets the Enactive Mind (Reprint edition). The MIT Press.
将社交媒体信息与音乐内容结合起来”。http://people.cs.uchicago.edu/~xiaofei/ACM-MM2010-bu.pdf。
Combining Social Media Information and Music Content”. http://people.cs.uchicago.edu/~xiaofei/ACM-MM2010-bu.pdf.
Connolly，D.（2003 年）。Web 架构图解：片段标识符。W3C。https://www.w3.org/2001/tag/fdesc54/slides.html
Connolly, D. (2003). Web Architecture Illustrated: Fragment Identifiers. W3C. https://www.w3.org/2001/tag/fdesc54/slides.html
Daga, E.、Peñuela AM、Motta, E. (2019)。在 RDF 中建模和查询列表。一项实用研究。第三届数据网络查询研讨会 (QuWeDa) 论文集。
Daga, E., Peñuela A. M., Motta, E. (2019). Modelling and Querying Lists in RDF. A Pragmatic Study. Proceedings of the 3rd Workshop on Querying the Web of Data (QuWeDa).
Dan McCreary。（2022 年 4 月 7 日）。个人知识图谱。面向数据科学。https://towardsdatascience.com/personal-knowledge-graphs-9a23a0b099af
Dan McCreary. (2022, April 7). Personal Knowledge Graphs. Towards Data Science. https://towardsdatascience.com/personal-knowledge-graphs-9a23a0b099af
以数据为中心的宣言。（nd）。2022 年 12 月 30 日检索自 http://datacentricmanifesto.org/
Data-Centric Manifesto. (n.d.). Retrieved December 30, 2022, from http://datacentricmanifesto.org/
De Jaegher, H. 和 Di Paolo, E. (2007)。参与式意义建构：一种社会认知的行为方法。现象学与认知科学，6(4)，485–507。https://doi.org/10.1007/s11097-007-9076-9
De Jaegher, H., & Di Paolo, E. (2007). Participatory sense-making: An enactive approach to social cognition. Phenomenology and the Cognitive Sciences, 6(4), 485–507. https://doi.org/10.1007/s11097-007-9076-9
De Mulder, G.、De Meester, B.、Heyvaert, P.、Taelman, R.、Dimou, A. 和 Verborgh, R. (2021)。 PROV4ITDaTa：将个人数据透明且直接地传输到个人存储。 695–697。 https://doi.org/10.1145/3442442.3458608
De Mulder, G., De Meester, B., Heyvaert, P., Taelman, R., Dimou, A., & Verborgh, R. (2021). PROV4ITDaTa: Transparent and direct transferof personal data to personal stores. 695–697. https://doi.org/10.1145/3442442.3458608
去中心化标识符 (DID) v1.0。（2022 年 7 月 19 日）。https://www.w3.org/TR/did-core/#production-0
Decentralized Identifiers (DIDs) v1.0. (2022, July 19). https://www.w3.org/TR/did-core/#production-0
Desruisseaux, B. (2009)。互联网日历和调度核心对象规范 (iCalendar)（征求意见 RFC 5545）。互联网工程任务组。https://doi.org/10.17487/RFC5545
Desruisseaux, B. (2009). Internet Calendaring and Scheduling Core Object Specification (iCalendar) (Request for Comments RFC 5545). Internet Engineering Task Force. https://doi.org/10.17487/RFC5545
Di Paolo, EA、Buhrmann, T. 和 Barandiaran, XE (2017)。感觉运动生活：一个实施提案（第一版）。牛津大学出版社。
Di Paolo, E. A., Buhrmann, T., & Barandiaran, X. E. (2017). Sensorimotor life: An enactive proposal (First edition). Oxford University Press.
Di Paolo, EA、Rohde, M. 和 De Jaegher, H. (2010)。《行动思维的视野：价值观、社交互动和游戏》。收录于 J. Stewart、O. Gapenne 和 EA Di Paolo (Eds.) 的《行动：迈向认知科学的新范式》(第 0 页)。麻省理工学院出版社。https://doi.org/10.7551/mitpress/9780262014601.003.0003
Di Paolo, E. A., Rohde, M., & De Jaegher, H. (2010). Horizons for the Enactive Mind: Values, Social Interaction, and Play. In J. Stewart, O. Gapenne, & E. A. Di Paolo (Eds.), Enaction: Toward a New Paradigm for Cognitive Science (p. 0). The MIT Press. https://doi.org/10.7551/mitpress/9780262014601.003.0003
DocQuery：由大型语言模型驱动的文档查询引擎。（2022 年）。[Python]。impira。https://github.com/impira/docquery（原创作品于 2022 年发表）
DocQuery: Document Query Engine Powered by Large Language Models. (2022). [Python]. impira. https://github.com/impira/docquery (Original work published 2022)
Dragan, L.、Handschuh, S. 和 Decker, S. (2011)。工作中的语义桌面：互连注释 (第 24 页)。https://doi.org/10.1145/2063518.2063521
Dragan, L., Handschuh, S., & Decker, S. (2011). The semantic desktop at work: Interlinking notes (p. 24). https://doi.org/10.1145/2063518.2063521
Edwards，B.（nd）。BeOS 是什么？为什么人们喜欢它？How-To Geek。2022 年 11 月 19 日检索自 https://www.howtogeek.com/696193/what-was-beos-and-why-did-people-love-it/
Edwards, B. (n.d.). What Was BeOS, and Why Did People Love It? How-To Geek. Retrieved November 19, 2022, from https://www.howtogeek.com/696193/what-was-beos-and-why-did-people-love-it/
Egbert, M. 和 Cañamero, L.（2014 年，8 月 1 日）。基于习惯的基本变量调节。 https://doi.org/10.7551/978-0-262-32621-6-ch029
Egbert, M., & Cañamero, L. (2014, August 1). Habit-Based Regulation of Essential Variables. https://doi.org/10.7551/978-0-262-32621-6-ch029
Esposito, E. (2022)。人工智能通信：算法如何产生社交智能。麻省理工学院出版社。
Esposito, E. (2022). Artificial Communication: How Algorithms Produce Social Intelligence. MIT Press.
FOAF 词汇规范。（2014 年 1 月 17 日）。https://web.archive.org/web/20220517235714/http://xmlns.com/foaf/spec/
FOAF Vocabulary Specification. (2014, January 17). https://web.archive.org/web/20220517235714/http://xmlns.com/foaf/spec/
foambubble, V. (2022)。泡沫。泡沫。https://foambubble.github.io/foam/
foambubble, V. (2022). Foam. Foam. https://foambubble.github.io/foam/
Foerster, HV 和 Poerksen, B. (2002)。《理解系统：关于认识论和伦理学的对话》。Kluwer Academic/Plenum Publishers。
Foerster, H. V., & Poerksen, B. (2002). Understanding Systems: Conversations on Epistemology and Ethics. Kluwer Academic/Plenum Publishers.
Fong, B. 和 Spivak, D. (2018)。“超图类别”。https://arxiv.org/abs/1806.08304。
Fong, B. and Spivak, D. (2018). “Hypergraph Categories”. https://arxiv.org/abs/1806.08304.
Forte, T. (2022)。构建第二大脑：一种组织数字生活和释放创造潜力的有效方法。Profile Books Ltd.
Forte, T. (2022). Building a second brain: A proven method to organise your digital life and unlock your creative potential. Profile Books Ltd.
Freire, P. (2014)。《被压迫者教育学》：30 周年纪念版。Bloomsbury Publishing。
Freire, P. (2014). Pedagogy of the Oppressed: 30th anniversary edition. Bloomsbury Publishing.
Gallagher, S. (2022)。《现象学》（第二版，2022 年版）。Palgrave Macmillan。
Gallagher, S. (2022). Phenomenology (2nd ed. 2022 edition). Palgrave Macmillan.
Garshol, L. (2003)。“与主题图和 RDF 共存”，https://ontopia.net/topicmaps/materials/tmrdf.html。
Garshol, L. (2003). “Living with topic maps and RDF”, https://ontopia.net/topicmaps/materials/tmrdf.html.
Giaro，M.（2022 年 2 月 21 日）。你记笔记的方式不对——原因如下。Medium。https://mattgiaro.medium.com/youre-taking-notes-the-wrong-way-here-s-why-89bd16ef6fc7
Giaro, M. (2022, February 21). You’re Taking Notes the WRONG Way—Here’s Why. Medium. https://mattgiaro.medium.com/youre-taking-notes-the-wrong-way-here-s-why-89bd16ef6fc7
Gifford, DK、Jouvelot, P.、Sheldon, MA 和 O'Toole, JW (1991)。语义文件系统。第十三届 ACM 操作系统原理研讨会论文集 – SOSP '91，第 16-25 页。https://doi.org/10.1145/121132.121138
Gifford, D. K., Jouvelot, P., Sheldon, M. A., & O’Toole, J. W. (1991). Semantic file systems. Proceedings of the Thirteenth ACM Symposium on Operating Systems Principles – SOSP ’91, 16–25. https://doi.org/10.1145/121132.121138
Glanville, R. (2007)。再试一次。再失败一次。失败得更好：设计中的控制论和控制论中的设计。Kybernetes，36(9/10)，1173–1206。https://doi.org/10.1108/03684920710827238
Glanville, R. (2007). Try again. Fail again. Fail better: The cybernetics in design and the design in cybernetics. Kybernetes, 36(9/10), 1173–1206. https://doi.org/10.1108/03684920710827238
Global Data Geeks（导演）。（2021 年 1 月 15 日）。什么是个人知识图谱 - 与 Ruben Verborgh 一起 — The Graph Show。https://www.youtube.com/watch?v=2EP35HO2HVQ
Global Data Geeks (Director). (2021, January 15). What is a Personal Knowledge Graph- with Ruben Verborgh—The Graph Show. https://www.youtube.com/watch?v=2EP35HO2HVQ
Google（导演）。（2007 年 2 月 13 日）。俳句：操作系统。https://www.youtube.com/watch?v=LxAQxGQB1A8
Google (Director). (2007, February 13). Haiku: The Operating System. https://www.youtube.com/watch?v=LxAQxGQB1A8
Gosnell, D. 和 Broecheler, M. (2020)。图表思维。《图表数据从业者指南》。https://www.oreilly.com/library/view/the-practitioners-guide/9781492044062/ch01.html
Gosnell, D., & Broecheler, M. (2020). Graph Thinking. In The Practitioner’s Guide to Graph Data. https://www.oreilly.com/library/view/the-practitioners-guide/9781492044062/ch01.html
Gruber, TR (1993)。可移植本体规范的翻译方法——将本体定义为规范。https://tomgruber.org/writing/ontolingua-kaj-1993
Gruber, T. R. (1993). A Translation Approach to Portable Ontology Specifications—Definition of Ontology as Specification. https://tomgruber.org/writing/ontolingua-kaj-1993
Gustafson, J. (2020)。“什么是分布式知识图谱？”。KFG 笔记。https://notes.knowledgefutures.org/pub/belji1gd。
Gustafson, J. (2020). “What is a Distributed Knowledge Graph?”. KFG Notes. https://notes.knowledgefutures.org/pub/belji1gd.
Gyrard, A.、Gaur, M.、Shekarpour, S.、Thirunarayan, K. 和 Sheth, A. (2018)。个性化健康知识图谱。7.
Gyrard, A., Gaur, M., Shekarpour, S., Thirunarayan, K., & Sheth, A. (2018). Personalized Health Knowledge Graph. 7.
Haiku, Inc.（2022 年）。Haiku 项目。Haiku 项目。https://www.haiku-os.org/
Haiku, Inc. (2022). Haiku Project. Haiku Project. https://www.haiku-os.org/
Haiku, Inc.（2022 年）。跟踪器。https://www.haiku-os.org/docs/userguide/en/tracker.html
Haiku, Inc. (2022). Tracker. https://www.haiku-os.org/docs/userguide/en/tracker.html
Haiku，V.（2022 年 9 月 12 日）。俳句书：消息传递基础。https://www.haiku-os.org/docs/api/app_messaging.html
Haiku, V. (2022, September 12). The Haiku Book: Messaging Foundations. https://www.haiku-os.org/docs/api/app_messaging.html
Halpin, H.、Clark, A. 和 Wheeler, M. (2013)。网络哲学：表征、实施、集体智慧。《哲学工程》（第 21-30 页）。John Wiley & Sons, Ltd. https://doi.org/10.1002/9781118700143.ch2
Halpin, H., Clark, A., & Wheeler, M. (2013). Philosophy of the Web: Representation, Enaction, Collective Intelligence. In Philosophical Engineering (pp. 21–30). John Wiley & Sons, Ltd. https://doi.org/10.1002/9781118700143.ch2
Hess, C. 和 Ostrom, E. (2019)。《将知识理解为公共资源——从理论到实践》。麻省理工学院出版社。
Hess, C., & Ostrom, E. (2019). Understanding knowledge as a commons—From theory to practice. The MIT Press.
Hogan, A.、Cochez, M. 和 Melo, G. de. (2022)。知识图谱。M & C，Morgan & Claypool 出版社。
Hogan, A., Cochez, M., & Melo, G. de. (2022). Knowledge graphs. M & C, Morgan & Claypool Publishers.
Hollister, S. (2022)。以数据为中心的宣言。http://www.datacentricmanifesto.org/
Hollister, S. (2022). Data-Centric Manifesto. http://www.datacentricmanifesto.org/
Horowitz，P.（2020 年 1 月 15 日）。Mac 上的便笺存储在哪里？OS X Daily。https://osxdaily.com/2020/01/15/where-notes-stored-locally-mac/
Horowitz, P. (2020, January 15). Where are Notes Stored on Mac? OS X Daily. https://osxdaily.com/2020/01/15/where-notes-stored-locally-mac/
https://nautil.us/will-ai-harm-us-better-to-ask-how-well-reckon-with-our-hybrid-nature-236098/
https://nautil.us/will-ai-harm-us-better-to-ask-how-well-reckon-with-our-hybrid-nature-236098/
Humdinger, B. (2009a)。属性（Haiku 用户指南）。https://www.haiku-os.org/docs/userguide/en/attributes.html
Humdinger, B. (2009a). Attributes (Haiku User Guide). https://www.haiku-os.org/docs/userguide/en/attributes.html
Humdinger, B. (2009b)。Haiku 文件类型（用户指南）。Haiku 基金会。https://www.haiku-os.org/docs/userguide/en/filetypes.html
Humdinger, B. (2009b). Haiku Filetypes (Userguide). The Haiku Foundation. https://www.haiku-os.org/docs/userguide/en/filetypes.html
Humdinger，B.（2019）。研讨会：文件类型、属性、索引和查询。https://www.haiku-os.org/docs/userguide/en/workshop-filetypes+attributes.html
Humdinger, B. (2019). Workshop: Filetypes, Attributes, Index and Queries. https://www.haiku-os.org/docs/userguide/en/workshop-filetypes+attributes.html
Hutchins, E. (1995)。《野外认知》。布拉德福德出版社。
Hutchins, E. (1995). Cognition in the Wild. Bradford Books.
Hutchins, E. (2010)。《行动、想象和洞察力》。J. Stewart、O. Gapenne 和 EA Di Paolo (Eds.) 著，《行动》（第 425-450 页）。麻省理工学院出版社。https://doi.org/10.7551/mitpress/9780262014601.003.0016
Hutchins, E. (2010). Enaction, Imagination, and Insight. In J. Stewart, O. Gapenne, & E. A. Di Paolo (Eds.), Enaction (pp. 425–450). The MIT Press. https://doi.org/10.7551/mitpress/9780262014601.003.0016
IANA，V.（2022）。MIME 类型（IANA 媒体类型）。Mozilla 基金会。https://developer.mozilla.org/en-US/docs/Web/HTTP/Basics_of_HTTP/MIME_types
IANA, V. (2022). MIME types (IANA media types). Mozilla Foundation. https://developer.mozilla.org/en-US/docs/Web/HTTP/Basics_of_HTTP/MIME_types
IPFS 为分布式网络提供支持。（2022 年）。https://ipfs.tech/
IPFS Powers the Distributed Web. (2022). https://ipfs.tech/
IPFS 规范。（2023 年）。[Makefile]。IPFS。https://github.com/ipfs/specs/blob/80db667971ef14e4e12249699e3cf6a898eca68c/UNIXFS.md（原作于 2015 年出版）
IPFS Specifications. (2023). [Makefile]. IPFS. https://github.com/ipfs/specs/blob/80db667971ef14e4e12249699e3cf6a898eca68c/UNIXFS.md (Original work published 2015)
IPFS，V.（2023 年 3 月 3 日）。什么是 IPFS？| IPFS 文档。https://docs.ipfs.tech/concepts/what-is-ipfs/
IPFS, V. (2023, March 3). What is IPFS? | IPFS Docs. https://docs.ipfs.tech/concepts/what-is-ipfs/
IPFS，V.（nd-a）。内容标识符 (CID) | IPFS 文档。2023 年 3 月 7 日检索自 https://docs.ipfs.tech/concepts/content-addressing/
IPFS, V. (n.d.-a). Content Identifiers (CIDs) | IPFS Docs. Retrieved March 7, 2023, from https://docs.ipfs.tech/concepts/content-addressing/
IPFS，V.（nd-b）。IPFS 的工作原理 | IPFS 文档。IPFS 文档。2022 年 11 月 15 日检索自 https://docs.ipfs.tech/concepts/how-ipfs-works/#content-addressing
IPFS, V. (n.d.-b). How IPFS works | IPFS Docs. IPFS Docs. Retrieved November 15, 2022, from https://docs.ipfs.tech/concepts/how-ipfs-works/#content-addressing
Jacobson, V.、Smetters, DK、Thornton, JD、Plass, M.、Briggs, N. 和 Braynard, R. (2012)。网络命名内容。Communications of the ACM，55(1)，117–124。https://doi.org/10.1145/2063176.2063204
Jacobson, V., Smetters, D. K., Thornton, J. D., Plass, M., Briggs, N., & Braynard, R. (2012). Networking named content. Communications of the ACM, 55(1), 117–124. https://doi.org/10.1145/2063176.2063204
Jones, W., Wenning, A., & Bruce, H. (2014)。人们如何重新查找文件、电子邮件和网页？https://hdl.handle.net/2142/47300
Jones, W., Wenning, A., & Bruce, H. (2014). How Do People Re-find Files, Emails and Web Pages? https://hdl.handle.net/2142/47300
Kalokyri, V.、Borgida, A. 和 Marian, A. (2018)。YourDigitalSelf：个人数字踪迹整合工具。第 27 届 ACM 国际信息与知识管理会议论文集，1963-1966 年。https://doi.org/10.1145/3269206.3269219
Kalokyri, V., Borgida, A., & Marian, A. (2018). YourDigitalSelf: A Personal Digital Trace Integration Tool. Proceedings of the 27th ACM International Conference on Information and Knowledge Management, 1963–1966. https://doi.org/10.1145/3269206.3269219
KDE eV (2023)。D-Bus 简介。开发者。https://develop.kde.org/docs/use/d-bus/introduction_to_dbus/
KDE e.V. (2023). Introduction to D-Bus. Developer. https://develop.kde.org/docs/use/d-bus/introduction_to_dbus/
KDE 的下一代语义搜索。（2014 年 2 月 24 日）。KDE.News。https://dot.kde.org/2014/02/24/kdes-next-generation-semantic-search
KDE’s Next Generation Semantic Search. (2014, February 24). KDE.News. https://dot.kde.org/2014/02/24/kdes-next-generation-semantic-search
Kernighan, BW 和 Plauger, PJ (1976)。软件工具。Addison-Wesley Pub. Co.
Kernighan, B. W., & Plauger, P. J. (1976). Software tools. Addison-Wesley Pub. Co.
Kim, G., Hong, T., Yim, M., Nam, J., Park, J., Yim, J., Hwang, W., Yun, S., Han, D., & Park, S. (2022)。无 OCR 文档理解转换器 (arXiv:2111.15664)。arXiv。http://arxiv.org/abs/2111.15664
Kim, G., Hong, T., Yim, M., Nam, J., Park, J., Yim, J., Hwang, W., Yun, S., Han, D., & Park, S. (2022). OCR-free Document Understanding Transformer (arXiv:2111.15664). arXiv. http://arxiv.org/abs/2111.15664
Krakauer, D. (2016)。“人工智能会伤害我们吗？最好问问自己如何面对我们的混合天性”。
Krakauer, D. (2016). “Will A.I. Harm Us? Better to Ask How We’ll Reckon With Our Hybrid Nature”.
Kuehn，M.（nd）。Niklas Luhmann 著《通过滑盒进行交流》。2022 年 8 月 11 日检索自 http://luhmann.surge.sh/communicating-with-slip-boxes
Kuehn, M. (n.d.). Communicating with Slip Boxes by Niklas Luhmann. Retrieved August 11, 2022, from http://luhmann.surge.sh/communicating-with-slip-boxes
Labs，P.（nd）。IPLD——内容可寻址网络的数据模型。IPLD。2022 年 12 月 25 日检索自 https://ipld.io/
Labs, P. (n.d.). IPLD – The data model of the content-addressable web. IPLD. Retrieved December 25, 2022, from https://ipld.io/
Lajos, G.（2019 年 6 月 30 日）。编织一个去中心化的（个人）知识语义网。https://doi.org/10.13140/RG.2.2.23855.07844
Lajos, G. (2019, June 30). Weaving a Decentralized Semantic Web of (Personal) Knowledge. https://doi.org/10.13140/RG.2.2.23855.07844
Li, X.、Tur, G.、Hakkani-Tur, D. 和 Li, Q. (2015)。对话理解中用户话语的个人知识图谱群体。2014 年 IEEE 口语技术研讨会，SLT 2014 – 论文集，224–229。https://doi.org/10.1109/SLT.2014.7078578
Li, X., Tur, G., Hakkani-Tur, D., & Li, Q. (2015). Personal knowledge graph population from user utterances in conversational understanding. 2014 IEEE Workshop on Spoken Language Technology, SLT 2014 – Proceedings, 224–229. https://doi.org/10.1109/SLT.2014.7078578
Linvega，Devine L. (2022)。话语。XXIVV。https://wiki.xxiivv.com/site/discourse.html
Linvega, Devine L. (2022). Discourse. XXIIVV. https://wiki.xxiivv.com/site/discourse.html
Litt, G. (2021)。带上自己的客户。https://www.geoffreylitt.com/2021/03/05/bring-your-own-client.html
Litt, G. (2021). Bring Your Own Client. https://www.geoffreylitt.com/2021/03/05/bring-your-own-client.html
LSoft Technologies Inc. (2022)。WinFS 概述—NTFS.com。WinFS 概述。https://www.ntfs.com/winfs_basics.htm
LSoft Technologies Inc. (2022). WinFS Overview—NTFS.com. WinFS Overview. https://www.ntfs.com/winfs_basics.htm
Ludington，J.（2022 年 10 月 25 日）。开发人员：是时候考虑将 IPFS 作为 HTTP 的替代方案了。新堆栈。https://thenewstack.io/devs-its-time-to-consider-ipfs-as-an-alternative-to-http/
Ludington, J. (2022, October 25). Devs: It’s Time to Consider IPFS as an Alternative to HTTP. The New Stack. https://thenewstack.io/devs-its-time-to-consider-ipfs-as-an-alternative-to-http/
卢曼，N.（1981）。 Zettelkästen 的通讯。载于 H. Baier、HM Kepplinger 和 K. Reumann（编），《Öffentliche Meinung und sozialer Wandel》/《公众舆论与社会变革》（第 222-228 页）。 VS Verlag für Sozialwissenschaften。 https://doi.org/10.1007/978-3-322-87749-9_19
Luhmann, N. (1981). Kommunikation mit Zettelkästen. In H. Baier, H. M. Kepplinger, & K. Reumann (Eds.), Öffentliche Meinung und sozialer Wandel / Public Opinion and Social Change (pp. 222–228). VS Verlag für Sozialwissenschaften. https://doi.org/10.1007/978-3-322-87749-9_19
Luhmann, N.、Baecker, D. 和 Barrett, R. (2018)。组织与决策。剑桥大学出版社。
Luhmann, N., Baecker, D., & Barrett, R. (2018). Organization and decision. Cambridge University Press.
Luhmann, N.、Baecker, D. 和 Barrett, R. (2018)。组织与决策。剑桥大学出版社。
Luhmann, N., Baecker, D., & Barrett, R. (2018). Organization and decision. Cambridge University Press.
Malcolm, N. (2004)。托马斯·哈里森和他的“研究方舟”——知识组织史上的一幕。《十七世纪》，19(2)，196–232。https://doi.org/10.1080/0268117X.2004.10555543
Malcolm, N. (2004). Thomas Harrison and his ‘Ark of Studies’ An Episode in the History of the Organization of Knowledge. The Seventeenth Century, 19(2), 196–232. https://doi.org/10.1080/0268117X.2004.10555543
Mather，Neil。（2021 年 12 月 29 日）。Node Club。Neil 的数字花园。https://commonplace.doubleloop.net/node-club
Mather, Neil. (2021, December 29). Node Club. Neil’s Digital Garden. https://commonplace.doubleloop.net/node-club
Maturana, HR 和 Varela, FJ (1980)。《自创生与认知：生命的实现》（第一版）。D. Reidel 出版公司。（原作出版于 1972 年）
Maturana, H. R., & Varela, F. J. (1980). Autopoiesis and Cognition: The Realization of the Living (1st edition). D. Reidel Publishing Company. (Original work published 1972)
Maturana, HR 和 Varela, FJ (1992)。《知识之树：人类理解的生物学根源》（修订版）。Shambhala；由兰登书屋在美国发行。
Maturana, H. R., & Varela, F. J. (1992). The tree of knowledge: The biological roots of human understanding (Rev. ed). Shambhala ; Distributed in the U.S. by Random House.
Matuschak, A. (nd)。常青笔记应以概念为导向。Andy 的工作笔记。检索日期：2023 年 4 月 22 日，来自 https://notes.andymatuschak.org/Evergreen_notes_should_be_concept-driven
Matuschak, A. (n.d.). Evergreen notes should be concept-oriented. Andyʼs Working Notes. Retrieved April 22, 2023, from https://notes.andymatuschak.org/Evergreen_notes_should_be_concept-oriented
Matuschak，A.（nd）。Evergreen Notes。安迪的工作笔记。2023 年 12 月 1 日检索自 https://notes.andymatuschak.org/Evergreen_notes
Matuschak, A. (n.d.). Evergreen Notes. Andy’s Working Notes. Retrieved December 1, 2023, from https://notes.andymatuschak.org/Evergreen_notes
Matuschak, A. & Nielsen, M. (2019) “我们如何开发变革性的思维工具？”，https://numinous.productions/ttft
Matuschak, A. & Nielsen, M. (2019) “How can we develop transformative tools for thought?”, https://numinous.productions/ttft
McComb, D. (2018)。软件荒原：以应用为中心的思维模式如何阻碍我们的企业。Technics Publications。
McComb, D. (2018). Software Wasteland: How the Application-Centric Mindset is Hobbling our Enterprises. Technics Publications.
McComb, D. (2019)。《以数据为中心的革命：恢复企业信息系统的理智》（第一版）。Technics Publications。
McComb, D. (2019). The Data-Centric Revolution: Restoring Sanity to Enterprise Information Systems (First edition). Technics Publications.
McNamee, R. (2019)。Zucked：醒来面对 Facebook 灾难。HarperCollins。
McNamee, R. (2019). Zucked: Waking Up to the Facebook Catastrophe. HarperCollins.
Melnikov, A. (2022)。IANA 媒体类型。互联网号码分配机构。https://www.iana.org/assignments/media-types/media-types.xhtml
Melnikov, A. (2022). IANA Media Types. Internet Assigned Numbers Authority. https://www.iana.org/assignments/media-types/media-types.xhtml
Melvær，K.（2022 年 2 月 18 日）。关于 Markdown 的想法。Smashing Magazine。https://www.smashingmagazine.com/2022/02/thoughts-on-markdown/
Melvær, K. (2022, February 18). Thoughts On Markdown. Smashing Magazine. https://www.smashingmagazine.com/2022/02/thoughts-on-markdown/
Milo，N.（2022 年 8 月 9 日）。MOCs 概述——LYT 套件。https://notes.linkingyourthinking.com/Cards/MOCs+Overview
Milo, N. (2022, August 9). MOCs Overview—LYT Kit. https://notes.linkingyourthinking.com/Cards/MOCs+Overview
Mingers, J. (2013)。《自我生产系统：自创生的含义和应用》。Springer。
Mingers, J. (2013). Self-Producing Systems: Implications and Applications of Autopoiesis. Springer.
Mohanaraj, A. 和 Laursen, EN (nd)。对话中实体链接到动态演变的个人知识图谱。28。
Mohanaraj, A., & Laursen, E. N. (n.d.). Entity Linking to Dynamically-Evolving Personal Knowledge Graphs in Conversations. 28.
Montoya, D.、Tanon, TP、Abiteboul, S.、Senellart, P. 和 Suchanek, FM (2018)。个人信息管理知识库。11.
Montoya, D., Tanon, T. P., Abiteboul, S., Senellart, P., & Suchanek, F. M. (2018). A Knowledge Base for Personal Information Management. 11.
MS-OLEDS：简介。（nd）。检索日期：2022 年 11 月 19 日，来自 https://learn.microsoft.com/en-us/openspecs/windows_protocols/ms-oleds/554a69b6-497a-4fef-a0fe-792d0dcf50fb
MS-OLEDS: Introduction. (n.d.). Retrieved November 19, 2022, from https://learn.microsoft.com/en-us/openspecs/windows_protocols/ms-oleds/554a69b6-497a-4fef-a0fe-792d0dcf50fb
Murphy，J.（2022 年 4 月 27 日）。个人知识管理是胡说八道。其他生活。https://www.otherlife.co/pkm/
Murphy, J. (2022, April 27). Personal Knowledge Management is Bullshit. Other Life. https://www.otherlife.co/pkm/
Nagarjuna, G.（2013 年 7 月 14 日）。GNOWSYS – GNU 项目 — 自由软件基金会。https://www.gnu.org/software/gnowsys/
Nagarjuna, G. (2013, July 14). GNOWSYS – GNU Project—Free Software Foundation. https://www.gnu.org/software/gnowsys/
Nathan, P. (2021 年 8 月 13 日)。图形思维。知识技术。https://medium.com/knowledge-technologies/graph-thinking-5e9c85134ab0
Nathan, P. (2021, August 13). Graph Thinking. Knowledge Technologies. https://medium.com/knowledge-technologies/graph-thinking-5e9c85134ab0
NeXT Computer, Inc, V.（nd）。NeXTstep ObjectLinks。2022 年 11 月 19 日检索自 http://www.nextcomputers.org/NeXTfiles/Docs/NeXTStep/3.3/nd/ReleaseNotes/ObjectLinks.htmld/index.html
NeXT Computer, Inc, V. (n.d.). NeXTstep ObjectLinks. Retrieved November 19, 2022, from http://www.nextcomputers.org/NeXTfiles/Docs/NeXTStep/3.3/nd/ReleaseNotes/ObjectLinks.htmld/index.html
Norman, D. (1991)。“认知人工制品”。https://hci.ucsd.edu/10/readings/norman_cognitiveartifacts.pdf
Norman, D. (1991). “Cognitive Artifacts”. https://hci.ucsd.edu/10/readings/norman_cognitiveartifacts.pdf
奥本瑙尔，A.（2021）。 LN 014：图形操作系统。 https://alexanderobenauer.com/labnotes/014
Obenauer, A. (2021). LN 014: The Graph OS. https://alexanderobenauer.com/labnotes/014
黑曜石。（2023 年）。https://obsidian.md/
Obsidian. (2023). https://obsidian.md/
Orlowski，A.（2002 年 3 月 29 日）。数据库上的窗口 - 由 BeOS 老手切片和切块。https://www.theregister.com/2002/03/29/windows_on_a_database_sliced/
Orlowski, A. (2002, March 29). Windows on a database – sliced and diced by BeOS vets. https://www.theregister.com/2002/03/29/windows_on_a_database_sliced/
Ostrom, E. (2009)。“超越市场和国家：复杂经济系统的多中心治理”。https://www.nobelprize.org/uploads/2018/06/ostrom_lecture.pdf
Ostrom, E. (2009). “Beyond markets and states: polycentric governance of complex economic systems”. https://www.nobelprize.org/uploads/2018/06/ostrom_lecture.pdf
Ostrom, E. (2015)。《治理公共资源》。剑桥大学出版社。
Ostrom, E. (2015). Governing the Commons. Cambridge University Press.
Palladino, C.、Kuczera, A. 和 Neill, I. (2020)。对“未经思考”进行建模。人文学科中的图形技术 - 2020 年论文集，第 22 页。
Palladino, C., Kuczera, A., & Neill, I. (2020). Modeling the ‘Unthought.’ Graph Technologies in the Humanities – Proceedings 2020, 22.
Paolo, EAD、Cuffari, EC 和 Jaegher, HD (2018)。《语言体：生命与语言之间的连续性》。麻省理工学院出版社。
Paolo, E. A. D., Cuffari, E. C., & Jaegher, H. D. (2018). Linguistic Bodies: The Continuity between Life and Language. The MIT Press.
Park, H. (2014)。Memex 及其他：Vannevar Bush 的 Memex 设计轨迹。https://www.semanticscholar.org/paper/Memex-and-Beyond%3A-A-Design-Trajectory-from-Vannevar-Park/685ebc9d30a310e5485729d4d59f71b20b47f1e5
Park, H. (2014). Memex and Beyond: A Design Trajectory from Vannevar Bush’s Memex. https://www.semanticscholar.org/paper/Memex-and-Beyond%3A-A-Design-Trajectory-from-Vannevar-Park/685ebc9d30a310e5485729d4d59f71b20b47f1e5
Parrillo, F. 和 Tschudin, C. (2021)。行星际文件系统上的固体。2021 年 IFIP 网络会议 (IFIP Networking)，1-6。https://doi.org/10.23919/IFIPNetworking52078.2021.9472772
Parrillo, F., & Tschudin, C. (2021). Solid over the Interplanetary File System. 2021 IFIP Networking Conference (IFIP Networking), 1–6. https://doi.org/10.23919/IFIPNetworking52078.2021.9472772
Pinheiro, E.（2020 年 1 月 9 日）。 BeOS：另一个宇宙的 Mac OS X。Hackaday。 https://hackaday.com/2020/01/09/beos-the-alternate-universes-mac-os-x/
Pinheiro, E. (2020, January 9). BeOS: The Alternate Universe’s Mac OS X. Hackaday. https://hackaday.com/2020/01/09/beos-the-alternate-universes-mac-os-x/
Revol, F. (2017 年 2 月 7 日)。Haiku，一个你仍然可以学习的桌面。https://archive.fosdem.org/2017/schedule/event/desktops_haiku_desktop_still_learn_from/attachments/slides/1826/export/events/attachments/desktops_haiku_desktop_still_learn_from/slides/1826/FR_FOSDEM2017_Haiku.pdf
Revol, F. (2017, February 7). Haiku, a desktop you can still learn from. https://archive.fosdem.org/2017/schedule/event/desktops_haiku_desktop_still_learn_from/attachments/slides/1826/export/events/attachments/desktops_haiku_desktop_still_learn_from/slides/1826/FR_FOSDEM2017_Haiku.pdf
Revol, F.（nd）。俳句，一个你仍可从中学习的桌面。19。
Revol, F. (n.d.). Haiku, a desktop you can still learn from. 19.
Robbins, P. 和 Aydede, M. (Eds.)。(2008)。《剑桥情境认知手册》。剑桥大学出版社。https://doi.org/10.1017/CBO9780511816826
Robbins, P., & Aydede, M. (Eds.). (2008). The Cambridge Handbook of Situated Cognition. Cambridge University Press. https://doi.org/10.1017/CBO9780511816826
Rosenauer, G. (2023)。SEN – 语义扩展。http://www.sen-labs.org/
Rosenauer, G. (2023). SEN – Semantic ExteNsions. http://www.sen-labs.org/
Safavi, T.、Fourney, A.、Sim, R.、Juraszek, M.、Williams, S.、Friend, N.、Koutra, D. 和 Bennett, PN (2020)。面向个人网络的活动发现。第 13 届网络搜索和数据挖掘国际会议论文集，第 492–500 页。https://doi.org/10.1145/3336191.3371828
Safavi, T., Fourney, A., Sim, R., Juraszek, M., Williams, S., Friend, N., Koutra, D., & Bennett, P. N. (2020). Toward Activity Discovery in the Personal Web. Proceedings of the 13th International Conference on Web Search and Data Mining, 492–500. https://doi.org/10.1145/3336191.3371828
Sauermann, L.、Kiesel, M.、Schumacher, K. 和 Bernardi, A. (2009)。语义桌面。载于 A. Blumauer 和 T. Pellegrini (Eds.)，《社交语义网》（第 337-362 页）。Springer Berlin Heidelberg。https://doi.org/10.1007/978-3-540-72216-8_17
Sauermann, L., Kiesel, M., Schumacher, K., & Bernardi, A. (2009). Semantic Desktop. In A. Blumauer & T. Pellegrini (Eds.), Social Semantic Web (pp. 337–362). Springer Berlin Heidelberg. https://doi.org/10.1007/978-3-540-72216-8_17
Sawday, J. 和 Rhodes, N. (Eds.)。(2000)。文艺复兴时期的计算机：印刷时代初期的知识技术（第一版）。劳特利奇。
Sawday, J., & Rhodes, N. (Eds.). (2000). The Renaissance Computer: Knowledge Technology in the First Age of Print (1st edition). Routledge.
schema.org，V.（2022a，3 月 17 日）。数据模型 - Schema.org。https://schema.org/docs/datamodel.html
schema.org, V. (2022a, March 17). Data Model—Schema.org. https://schema.org/docs/datamodel.html
schema.org，V.（2022b，3 月 17 日）。Schema.org—Schemas—Schema.org。https://schema.org/docs/schemas.html
schema.org, V. (2022b, March 17). Schema.org—Schemas—Schema.org. https://schema.org/docs/schemas.html
Schepers, D.（2007 年 12 月 13 日）。交出钥匙：用户控制的元数据。https://www.w3.org/2007/08/video/positions/Schepers.html
Schepers, D. (2007, December 13). Handing Over the Keys: User-Controlled Metadata. https://www.w3.org/2007/08/video/positions/Schepers.html
Schmidt, JFK (2018)。Niklas Luhmann 的卡片索引：意外发现的制造。Sociologica，第 12 卷，53-60 页。https://doi.org/10.6092/ISSN.1971-8853/8350
Schmidt, J. F. K. (2018). Niklas Luhmann’s Card Index: The Fabrication of Serendipity. Sociologica, Vol 12, 53-60 Pages. https://doi.org/10.6092/ISSN.1971-8853/8350
Seidl, D. (2010)。第 7 章 组织科学与组织实践之间的生产性误解：从 Niklas Luhmann 的自创生系统理论看科学与实践的关系。收录于 R. Magalhães 和 R. Sanchez (Eds.) 编著的《管理高级系列》（第 6 卷，第 133-148 页）。Emerald Group Publishing Limited。https://doi.org/10.1108/S1877-6361(2009)0000006008
Seidl, D. (2010). Chapter 7 Productive Misunderstandings between Organization Science and Organization Practice: The Science–Practice Relation from the Perspective of Niklas Luhmann’s Theory of Autopoietic Systems. In R. Magalhães & R. Sanchez (Eds.), Advanced Series in Management (Vol. 6, pp. 133–148). Emerald Group Publishing Limited. https://doi.org/10.1108/S1877-6361(2009)0000006008
Seidl, D. 和 Becker, K.-H. (2005)。Niklas Luhmann 与组织研究。https://doi.org/10.5167/uzh-73446
Seidl, D., & Becker, K.-H. (2005). Niklas Luhmann and Organization Studies. https://doi.org/10.5167/uzh-73446
Seneviratne, O.、Harris, J.、Chen, C.-H. 和 McGuinness, DL (2021)。临床相关饮食建议的个人健康知识图谱 (arXiv:2110.10131)。arXiv。https://doi.org/10.48550/arXiv.2110.10131
Seneviratne, O., Harris, J., Chen, C.-H., & McGuinness, D. L. (2021). Personal Health Knowledge Graph for Clinically Relevant Diet Recommendations (arXiv:2110.10131). arXiv. https://doi.org/10.48550/arXiv.2110.10131
Siegel，D.（2009）。《拉动：语义网的力量将改变你的业务》（第 1 版）。精装版。
Siegel, D. (2009). Pull: The Power of the Semantic Web to Transform Your Business (1st edition). Portfolio Hardcover.
Siegel，D.（2016 年 3 月 23 日）。个人数据储物柜视觉高清版。[视频]。YouTube。https://www.youtube.com/watch?v=xOch5o3MhUg
Siegel, D. (2016, March 23). Personal Data Locker Vision HD. [Video]. YouTube. https://www.youtube.com/watch?v=xOch5o3MhUg
Silverston, L. (2020 年 11 月 18 日)。禅与数据维护的艺术：数据“挖掘”和通用数据语义。数据管理通讯。https://tdan.com/zen-and-the-art-of-data-maintenance-data-mineing-and-universal-data-semantics/27543
Silverston, L. (2020, November 18). Zen and the Art of Data Maintenance: Data ‘Mine’ing and Universal Data Semantics. The Data Administration Newsletter. https://tdan.com/zen-and-the-art-of-data-maintenance-data-mineing-and-universal-data-semantics/27543
Singer-Vine，J.（2022 年）。Pdfplumber [Python]。https://github.com/jsvine/pdfplumber（原作于 2015 年出版）
Singer-Vine, J. (2022). Pdfplumber [Python]. https://github.com/jsvine/pdfplumber (Original work published 2015)
Singer, P. (2011)。《扩展的循环：伦理、进化和道德进步》。普林斯顿大学出版社。
Singer, P. (2011). The expanding circle: Ethics, evolution, and moral progress. Princeton University Press.
Singhal，A.（2012 年 5 月 16 日）。知识图谱简介：事物，而非字符串。谷歌。https://blog.google/products/search/introducing-knowledge-graph-things-not/
Singhal, A. (2012, May 16). Introducing the Knowledge Graph: Things, not strings. Google. https://blog.google/products/search/introducing-knowledge-graph-things-not/
Skjæveland, M. (2021)。OTTR 模板库。基于模式的本体工程进展，51。
Skjæveland, M. (2021). The OTTR Template Library. Advances in Pattern-Based Ontology Engineering, 51.
Skjæveland, MG、Balog, K.、Bernard, N.、Lajewska, W. 和 Linjordet, T. (2023)。个人知识图谱生态系统：调查和研究路线图 (arXiv:2304.09572)。arXiv。https://doi.org/10.48550/arXiv.2304.09572
Skjæveland, M. G., Balog, K., Bernard, N., Lajewska, W., & Linjordet, T. (2023). An Ecosystem for Personal Knowledge Graphs: A Survey and Research Roadmap (arXiv:2304.09572). arXiv. https://doi.org/10.48550/arXiv.2304.09572
Smart, PR (2013)。《网络扩展的思维》。《哲学工程》（第 116-133 页）。John Wiley & Sons, Ltd. https://doi.org/10.1002/9781118700143.ch8
Smart, P. R. (2013). The Web-Extended Mind. In Philosophical Engineering (pp. 116–133). John Wiley & Sons, Ltd. https://doi.org/10.1002/9781118700143.ch8
Smart, PR、Engelbrecht, PC、Braines, D.、Strub, M. 和 Giammanco, C. (2010)。网络扩展思维。《军事联盟行动的网络科学：信息交换与互动》（第 191-236 页）。IGI Global。https://doi.org/10.4018/978-1-61520-855-5.ch010
Smart, P. R., Engelbrecht, P. C., Braines, D., Strub, M., & Giammanco, C. (2010). The Network-Extended Mind. In Network Science for Military Coalition Operations: Information Exchange and Interaction (pp. 191–236). IGI Global. https://doi.org/10.4018/978-1-61520-855-5.ch010
Smolen, Paul；Zhang, Yili；Byrne, John H. (2016)。“学习的正确时间：间隔学习的机制和优化”。《自然神经科学评论》。17 (2): 77–88。arXiv:1606.08370。Bibcode：2016arXiv160608370S。doi：10.1038/nrn.2015.18。PMC 5126970。PMID 26806627。
Smolen, Paul; Zhang, Yili; Byrne, John H. (2016). “The right time to learn: mechanisms and optimization of spaced learning”. Nature Reviews Neuroscience. 17 (2): 77–88. arXiv:1606.08370. Bibcode:2016arXiv160608370S. doi:10.1038/nrn.2015.18. PMC 5126970. PMID 26806627.
社交电视和数据的未来。（2022 年 10 月 26 日）。BBC 研发部。https://www.bbc.co.uk/rd/blog/2022-10-social-tv-and-personal-data
Social TV and the future of data. (2022, October 26). BBC R&D. https://www.bbc.co.uk/rd/blog/2022-10-social-tv-and-personal-data
solidproject。（2023 年）。固体。https://solidproject.org/about
solidproject. (2023). Solid. https://solidproject.org/about
Spencer-Brown, G. (1979)。形式法则。达顿。
Spencer-Brown, G. (1979). Laws of form. Dutton.
Sporny, M.（2020 年 7 月 16 日）。 JSON-LD 1.1。 https://www.w3.org/TR/json-ld/
Sporny, M. (2020, July 16). JSON-LD 1.1. https://www.w3.org/TR/json-ld/
2021 年 Stack Overflow 开发者调查。（2021 年）。Stack Overflow。https://insights.stackoverflow.com/survey/2021
Stack Overflow Developer Survey 2021. (2021). Stack Overflow. https://insights.stackoverflow.com/survey/2021
Stewart, J.、Gapenne, O. 和 Paolo, EAD（编）。（2014 年）。《Enaction：迈向认知科学的新范式》（插图版）。布拉德福德出版社。
Stewart, J., Gapenne, O., & Paolo, E. A. D. (Eds.). (2014). Enaction: Toward a New Paradigm for Cognitive Science (Illustrated edition). Bradford Books.
Tana – 一切操作系统。 （2022 年 10 月 4 日）。塔纳。 https://tana-website-pvfdtog40-tana.vercel.app/
Tana – The Everything OS. (2022, October 4). Tana. https://tana-website-pvfdtog40-tana.vercel.app/
Teevan, J.、Alvarado, C.、Ackerman, MS 和 Karger, DR (2004)。完美的搜索引擎还不够：定向搜索中的定向越野行为研究。415–422。
Teevan, J., Alvarado, C., Ackerman, M. S., & Karger, D. R. (2004). The perfect search engine is not enough: A study of orienteering behavior in directed search. 415–422.
Telburt，J.（2022 年 2 月 16 日）。数据不言自明：数据乱扔垃圾。数据管理通讯。https://tdan.com/data-speaks-for-itself-data-littering/29122
Telburt, J. (2022, February 16). Data Speaks for Itself: Data Littering. The Data Administration Newsletter. https://tdan.com/data-speaks-for-itself-data-littering/29122
佛兰德斯政府和 Solid：“佛兰德斯历史上的重要里程碑。”（2020 年）。https://www.inrupt.com/blog/flanders-solid
The Flanders Government and Solid: “An important milestone in Flemish history.” (2020). https://www.inrupt.com/blog/flanders-solid
Thierer, A. (2016)。无许可创新：全面技术自由的持续案例。乔治梅森大学梅卡图斯中心。
Thierer, A. (2016). Permissionless Innovation: The Continuing Case for Comprehensive Technological Freedom. Mercatus Center at George Mason University.
Thompson, E. (2007)。《生活中的心灵：生物学、现象学和心灵科学》。哈佛大学出版社贝尔纳普出版社。
Thompson, E. (2007). Mind in life: Biology, phenomenology, and the sciences of mind. Belknap Press of Harvard University Press.
Tigunova, A.、Yates, A.、Mirza, P. 和 Weikum, G. (2020)。CHARM：从对话中推断个人属性。2020 年自然语言处理实证方法会议 (EMNLP) 论文集，5391–5404。https://doi.org/10.18653/v1/2020.emnlp-main.434
Tigunova, A., Yates, A., Mirza, P., & Weikum, G. (2020). CHARM: Inferring Personal Attributes from Conversations. Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), 5391–5404. https://doi.org/10.18653/v1/2020.emnlp-main.434
Torbati, GH、Yates, A. 和 Weikum, G. (2021)。聊天所得：使用对话个性化基于搜索的推荐 (arXiv:2109.04716)。arXiv。https://doi.org/10.48550/arXiv.2109.04716
Torbati, G. H., Yates, A., & Weikum, G. (2021). You Get What You Chat: Using Conversations to Personalize Search-based Recommendations (arXiv:2109.04716). arXiv. https://doi.org/10.48550/arXiv.2109.04716
Usher, G., & Whitty, J. (2017)。项目管理阴阳：将项目成功与客户满意度结合起来。项目管理研究与实践，第 4 期。https://doi.org/10.5130/pmrp.v4i0.5523
Usher, G., & Whitty, J. (2017). Project Management Yinyang: Coupling project success and client satisfaction. Project Management Research and Practice, 4. https://doi.org/10.5130/pmrp.v4i0.5523
Vahlo, J. (2017)。电子游戏玩法自主性的动态描述。游戏研究，17(1)。https://gamestudies.org/1701/articles/vahlo
Vahlo, J. (2017). An Enactive Account of the Autonomy of Videogame Gameplay. Game Studies, 17(1). https://gamestudies.org/1701/articles/vahlo
Vander Wal, T.（2007 年 2 月 2 日）。大众分类法的创造和定义。https://vanderwal.net/folksonomy.html
Vander Wal, T. (2007, February 2). Folksonomy Coinage and Definition. https://vanderwal.net/folksonomy.html
Vannur, LS、Ganesan, B.、Nagalapatti, L.、Patel, H. 和 Thippeswamy, MN (2020)。个人知识库人口的数据增强 (arXiv:2002.10943)。arXiv。https://doi.org/10.48550/arXiv.2002.10943
Vannur, L. S., Ganesan, B., Nagalapatti, L., Patel, H., & Thippeswamy, M. N. (2020). Data Augmentation for Personal Knowledge Base Population (arXiv:2002.10943). arXiv. https://doi.org/10.48550/arXiv.2002.10943
Varela, F. (1987)。在行走中铺就道路：一位生物学家眼中的新生物学。https://www.semanticscholar.org/paper/Laying-down-a-path-in-walking%3A-A-biologist%E2%80%99s-look-a-Varela/495fc912cb6ef16da41a75c7385f5a5a1fc5db3d
Varela, F. (1987). Laying down a path in walking: A biologist’s look at a new biology. https://www.semanticscholar.org/paper/Laying-down-a-path-in-walking%3A-A-biologist%E2%80%99s-look-a-Varela/495fc912cb6ef16da41a75c7385f5a5a1fc5db3d
Varela, FJ、Thompson, E. 和 Rosch, E. (1991)。具身心智：认知科学与人类经验。麻省理工学院出版社。
Varela, F. J., Thompson, E., & Rosch, E. (1991). The embodied mind: Cognitive science and human experience. MIT Press.
Vef, M.-A.、Steiner, R.、Salkhordeh, R.、Steinkamp, J.、Vennetier, F.、Smigielski, J.-F. 和 Brinkmann, A. (2020)。DelveFS – 用于对象存储的事件驱动语义文件系统。2020 年 IEEE 国际集群计算会议 (CLUSTER)，第 35–46 页。https://doi.org/10.1109/CLUSTER49012.2020.00014
Vef, M.-A., Steiner, R., Salkhordeh, R., Steinkamp, J., Vennetier, F., Smigielski, J.-F., & Brinkmann, A. (2020). DelveFS – An Event-Driven Semantic File System for Object Stores. 2020 IEEE International Conference on Cluster Computing (CLUSTER), 35–46. https://doi.org/10.1109/CLUSTER49012.2020.00014
Velitchkov，I.（2017 年 5 月 11 日）。我们还崇拜知识金字塔吗？https://www.strategicstructures.com/?p=1077
Velitchkov, I. (2017, May 11). Do We Still Worship The Knowledge Pyramid? https://www.strategicstructures.com/?p=1077
Velitchkov, I. (2020)。基本平衡：停止观察，开始了解组织运作的原理。Kvistgaard。
Velitchkov, I. (2020). Essential Balances: Stop Looking and Start Seeing What Makes Organizations Work. Kvistgaard.
Velitchkov，I.（2021a，4 月 4 日）。链接数据吸收。https://www.strategicstructures.com/?p=2193
Velitchkov, I. (2021a, April 4). Linked Data uptake. https://www.strategicstructures.com/?p=2193
Velitchkov，I.（主任）。（2021b，10 月 17 日）。使用知识图谱实现更好的项目管理。https://www.youtube.com/watch?v=AZmhGV7ypQ4
Velitchkov, I. (Director). (2021b, October 17). Better Project Management with Knowledge Graphs. https://www.youtube.com/watch?v=AZmhGV7ypQ4
Verborgh，R.（2020 年 12 月 7 日）。数据生态系统促进可持续创新。https://ruben.verborgh.org/blog/2020/12/07/a-data-ecosystem-fosters-sustainable-innovation/
Verborgh, R. (2020, December 7). A data ecosystem fosters sustainable innovation. https://ruben.verborgh.org/blog/2020/12/07/a-data-ecosystem-fosters-sustainable-innovation/
Virgo, N. (2011).热力学和生命系统的结构。
Virgo, N. (2011). Thermodynamics and the structure of living systems.
von Foerster, H. (2003)。记忆是什么？它也可能具有后见之明和先见之明？在 H. von Foerster (编辑) 的《理解理解：关于控制论和认知的论文集》（第 101-131 页）中。Springer。https://doi.org/10.1007/0-387-21722-3_3
von Foerster, H. (2003). What Is Memory that It May Have Hindsight and Foresight as well? In H. von Foerster (Ed.), Understanding Understanding: Essays on Cybernetics and Cognition (pp. 101–131). Springer. https://doi.org/10.1007/0-387-21722-3_3
W3C，V.（2015 年）。语义网，链接数据 — W3C。https://www.w3.org/standards/semanticweb/data
W3C, V. (2015). Semantic Web, Linked Data—W3C. https://www.w3.org/standards/semanticweb/data
W3C。（2017a，2 月 23 日）。Web 注释数据模型。https://www.w3.org/TR/annotation-model/
W3C. (2017a, February 23). Web Annotation Data Model. https://www.w3.org/TR/annotation-model/
W3C。（2017b，7 月 20 日）。形状约束语言 (SHACL)。https://www.w3.org/TR/shacl/
W3C. (2017b, July 20). Shapes Constraint Language (SHACL). https://www.w3.org/TR/shacl/
W3C。（nd）。WebAnnotation 主页。WebAnnotation：2022 年 9 月 21 日检索自 http://webannotation.org/
W3C. (n.d.). WebAnnotation Home. WebAnnotation: Retrieved September 21, 2022, from http://webannotation.org/
Watson, JD (1968)。双螺旋：DNA 结构发现的个人记录。Weidenfeld 和 Nicolson。
Watson, J. D. (1968). The double helix: A personal account of the discovery of the structure of DNA. Weidenfeld and Nicolson.
为什么我使用 SHACL 来定义本体模型。（2021 年 6 月 1 日）。TopQuadrant, Inc. https://archive.topquadrant.com/shacl-blog/
Why I Use SHACL For Defining Ontology Models. (2021, June 1). TopQuadrant, Inc. https://archive.topquadrant.com/shacl-blog/
互联网为何需要星际文件系统。（2022 年 10 月 9 日）。IEEE Spectrum。https://spectrum.ieee.org/peer-to-peer-network
Why the Internet needs the InterPlanetary File System. (2022, October 9). IEEE Spectrum. https://spectrum.ieee.org/peer-to-peer-network
维德默，汉斯. （1983）。波洛波洛（第三版）。 https://theanarchistlibrary.org/library/pm-bolo-bolo.pdf
Widmer., Hans. (1983). Bolo’bolo (3rd ed.). https://theanarchistlibrary.org/library/p-m-bolo-bolo.pdf
维基数据。 （2019 年 12 月 30 日）。 https://www.wikidata.org/wiki/Wikidata:Main_Page
Wikidata. (2019, December 30). https://www.wikidata.org/wiki/Wikidata:Main_Page
Wilkinson, MD, Dumontier, M., Aalbersberg, Ij. J., Appleton, G., Axton, M., Baak, A., Blomberg, N., Boiten, J.-W., da Silva Santos, LB, Bourne, PE, Bouwman, J., Brookes, AJ, Clark, T., Crosas, M., Dillo, I., Dumon, O., Edmunds, S., Evelo, CT, Finkers, R., … Mons, B. (2016)。《FAIR 科学数据管理和管理指导原则》。《科学数据》，3(1)，160018。https://doi.org/10.1038/sdata.2016.18
Wilkinson, M. D., Dumontier, M., Aalbersberg, Ij. J., Appleton, G., Axton, M., Baak, A., Blomberg, N., Boiten, J.-W., da Silva Santos, L. B., Bourne, P. E., Bouwman, J., Brookes, A. J., Clark, T., Crosas, M., Dillo, I., Dumon, O., Edmunds, S., Evelo, C. T., Finkers, R., … Mons, B. (2016). The FAIR Guiding Principles for scientific data management and stewardship. Scientific Data, 3(1), 160018. https://doi.org/10.1038/sdata.2016.18
Worg。（2023 年 3 月 23 日）。你好，Worg，Org-Mode 社区！https://orgmode.org/worg/
Worg. (2023, March 23). Hello Worg, the Org-Mode Community! https://orgmode.org/worg/
Yu, D.、Sun, K.、Cardie, C. 和 Yu, D. (2020)。基于对话的关系提取。计算语言学协会第 58 届年会论文集，4927–4940。https://doi.org/10.18653/v1/2020.acl-main.444
Yu, D., Sun, K., Cardie, C., & Yu, D. (2020). Dialogue-Based Relation Extraction. Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, 4927–4940. https://doi.org/10.18653/v1/2020.acl-main.444
Zichichi, M.、Ferretti, S. 和 D'Angelo, G. (2020)。关于个人信息管理系统去中心化文件存储的效率 (arXiv:2007.03505)。arXiv。http://arxiv.org/abs/2007.03505
Zichichi, M., Ferretti, S., & D’Angelo, G. (2020). On the Efficiency of Decentralized File Storage for Personal Information Management Systems (arXiv:2007.03505). arXiv. http://arxiv.org/abs/2007.03505
Zuboff，S.（2019）。《监控资本主义时代：在权力新前沿为人类未来而战》（平装版）。简介书籍。
Zuboff, S. (2019). The age of surveillance capitalism: The fight for a human future at the new frontier of power (Paperback edition). Profile Books.